Speech intelligibility enhancement system and method

CROSS-REFERENCE TO RELATED APPLICATIONS

0001This application claims priority to U.S. Provisional Patent Application No. 61/052,553, filed May 12, 2008, the entirety of which is incorporated by reference herein.

BACKGROUND OF THE INVENTION

00021. Field of the Invention

0003The present invention generally relates to communication systems used to transmit speech signals. More particularly, the present invention relates to methods for enhancing the intelligibility of speech signals received over a communication network from a far-end telephony terminal for playback at a near-end telephony terminal.

00042. Background

0005Various background concepts will now be discussed in reference to an example conventional communication system <b>100</b> shown in <figref idref="DRAWINGS">FIG. 1</figref>. Communication system <b>100</b> includes a first telephony terminal <b>102</b> and a second telephony terminal <b>104</b> that are communicatively connected to each other via one or more communication network(s) <b>106</b>. For the purposes of this example, first telephony terminal <b>102</b> will be referred to as the “near end” of the network connection and second telephony terminal <b>104</b> will be referred to as the “far end” of the network connection. Each telephony terminal may comprise a telephony device, such as a corded telephone, cordless telephone, cellular telephone or Bluetooth® headset.

0006First telephony terminal <b>102</b> operates in a well-known manner to pick up speech signals representing the voice of a near-end user <b>108</b> via a microphone <b>114</b> and to transmit such speech signals over network(s) <b>106</b> to second telephony terminal <b>104</b>. Second telephony terminal <b>104</b> operates in a well-known manner to play back the received speech signals to a far-end user <b>110</b> via a loudspeaker <b>118</b>. Conversely, second telephony terminal <b>104</b> operates in a well-known manner to pick up speech signals representing the voice of far-end user <b>110</b> via a microphone <b>116</b> and to transmit such speech signals over network(s) <b>106</b> to first telephony terminal <b>102</b>. First telephony terminal <b>102</b> operates in a well-known manner to play back the received speech signals to near-end user <b>108</b> via a loudspeaker <b>112</b>.

0007As further shown in <figref idref="DRAWINGS">FIG. 1</figref>, near-end user <b>108</b> is using first telephony terminal <b>102</b> in an environment that is subject to acoustic background noise. When this acoustic background noise becomes too loud, near-end user <b>108</b> may find the voice of far-end user <b>110</b> difficult to understand. This is because such loud acoustic background noise will tend to mask or drown out the voice of far-end user <b>110</b> that is being played back through loudspeaker <b>112</b> of first telephony terminal <b>102</b>. When this occurs, the natural response of near-end user <b>108</b> may be to adjust the volume of loudspeaker <b>112</b> (assuming that first telephony terminal <b>102</b> includes a volume control button or some other volume control means) so that the volume of the voice of far-end user <b>110</b> is increased. However, it is inconvenient for near-end user <b>108</b> to have to manually adjust the volume in this manner; it would be far more convenient if first telephony terminal <b>102</b> could automatically adjust the volume to the appropriate level in response to an increase in acoustic background noise.

0008Furthermore, although near-end user <b>108</b> may increase the volume of loudspeaker <b>112</b>, there is typically a limit on how much amplification can be applied to the speech signal received from far-end user <b>110</b> before that signal is subject to digital saturation or clipping. Additionally, even when the speech signal received from far-end user <b>110</b> has been amplified to a level immediately below which clipping occurs or to a level at which slight clipping occurs, the speech signal may still not be loud enough to be intelligible over the acoustic background noise.

0009Various techniques have been described in the literature that can be used to increase the loudness of a speech signal subject to a magnitude limit (such as amplitude compression) or to make the speech signal more intelligible. However, many of these techniques distort the speech signal.

0010What is needed, therefore, is a speech intelligibility enhancement (SIE) system and method that improves the intelligibility of a speech signal received over a communication network from a far-end telephony terminal for playback at a near-end telephony terminal when the near-end terminal is located in an environment with loud acoustic background noise. The desired SIE system and method should function automatically without any user input and also achieve improved intelligibility while minimizing distortion to the received speech signal.

BRIEF SUMMARY OF THE INVENTION

0011A speech intelligibility enhancement (SIE) system and method is described herein that may be used to improve the intelligibility of a speech signal to be played back by an audio device when the audio device is located in an environment with loud acoustic background noise. In an embodiment, the audio device comprises a near-end telephony terminal and the speech signal comprises a speech signal received over a communication network from a far-end telephony terminal for playback at the near-end telephony terminal.

0012In particular, a method for processing a portion of a speech signal to be played back by an audio device is described herein. In accordance with the method, a level of the speech signal is estimated. A level of background noise is estimated. A signal-to-noise ratio (SNR) is calculated based on the estimated level of the speech signal and the estimated level of the background noise. An amount of gain to be applied to the portion of the speech signal is calculated based on at least a difference between a predetermined SNR and the calculated SNR. The amount of gain is then applied to the portion of the speech signal.

0013A system for processing a portion of a speech signal to be played back by an audio device is also described herein. The system includes a level estimator and a logic block. The level estimator is configured to estimate a level of the speech signal. The logic block is configured to receive an estimated level of background noise, to calculate an SNR based on the estimated level of the speech signal and the estimated level of the background noise, to calculate an amount of gain to be applied to the portion of the speech signal based on at least a difference between a predetermined SNR and the calculated SNR and to apply the amount of gain to the portion of the speech signal.

0014An additional method for processing a speech signal to be played back by an audio device is described herein. In accordance with the method, a level of background noise is estimated. A linear gain is applied to the speech signal if a function of at least the estimated level of the background noise meets a first condition. A linear gain and compression are applied to the speech signal if the function of at least the estimated level of the background noise meets a second condition.

0015A further method for processing a speech signal to be played back by an audio device is described herein. In accordance with the method, a level of background noise is estimated. A linear gain is applied to the speech signal if a function of at least the estimated level of the background noise meets a first condition. A linear gain and dispersion filtering is applied to the speech signal if the function of least the estimated level of the background noise meets a second condition.

0016A still further method for processing a speech signal to be played back by an audio device is described herein. In accordance with the method, a level of background noise is estimated. A linear gain is applied to the speech signal if a function of at least the estimated level of the background noise meets a first condition. A linear gain and spectral shaping is applied to the speech signal if the function of at least the estimated level of the background noise meets a second condition.

0017Further features and advantages of the invention, as well as the structure and operation of various embodiments of the invention, are described in detail below with reference to the accompanying drawings. It is noted that the invention is not limited to the specific embodiments described herein. Such embodiments are presented herein for illustrative purposes only. Additional embodiments will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein.

BRIEF DESCRIPTION OF THE DRAWINGS/FIGURES

0018The accompanying drawings, which are incorporated herein and form part of the specification, illustrate the present invention and, together with the description, further serve to explain the principles of the invention and to enable a person skilled in the relevant art(s) to make and use the invention.

0019<figref idref="DRAWINGS">FIG. 1</figref> is a block diagram of an example conventional communication system.

0020<figref idref="DRAWINGS">FIG. 2</figref> is a block diagram of an example speech intelligibility enhancement (SIE) system in accordance with an embodiment of the present invention.

0021<figref idref="DRAWINGS">FIG. 3</figref> depicts a block diagram of a high-pass spectral shaping filter that may be used to implement an SIE system in accordance with an embodiment of the present invention.

0022<figref idref="DRAWINGS">FIG. 4</figref> is a graph showing a family of frequency response curves for a slowly-evolving spectral shaping filter in accordance with an embodiment of the present invention.

0023<figref idref="DRAWINGS">FIG. 5</figref> is a graph showing a family of frequency response curves for a rapidly-evolving spectral shaping filter in accordance with an embodiment of the present invention.

0024<figref idref="DRAWINGS">FIG. 6</figref> depicts a first plot that shows an example male speech waveform before dispersion filtering and a second plot that shows the same segment of speech waveform after dispersion filtering.

0025<figref idref="DRAWINGS">FIG. 7</figref> depicts a first plot that shows an example female speech waveform before dispersion filtering and a second plot that shows the same segment of speech waveform after dispersion filtering.

0026<figref idref="DRAWINGS">FIG. 8</figref> is a block diagram of an automatic gain control (AGC)/automatic volume boost (AVB)/compression block in accordance with an embodiment of the present invention.

0027<figref idref="DRAWINGS">FIG. 9</figref> is a block diagram of an example SIE system in accordance with an alternate embodiment of the present invention.

0028<figref idref="DRAWINGS">FIG. 10</figref> is a block diagram of AGC logic that may be used to implement in an SIE system in accordance with an alternate embodiment of the present invention.

0029<figref idref="DRAWINGS">FIG. 11</figref> is a block diagram that shows a telephony terminal in which an SIE system in accordance with an embodiment of the present invention is integrated with a sub-band acoustic canceller.

0030<figref idref="DRAWINGS">FIG. 12</figref> is a block diagram that shows an alternate telephony terminal in which an SIE system in accordance with an embodiment of the present invention is integrated with a sub-band acoustic canceller.

0031<figref idref="DRAWINGS">FIGS. 13-18</figref> depict flowcharts of various methods for processing a portion of a speech signal to be played back by an audio device in accordance with embodiments of the present invention.

0032<figref idref="DRAWINGS">FIG. 19</figref> depicts a flowchart of a method for operating an integrated speech intelligibility enhancement system and acoustic echo canceller in accordance with an embodiment of the present invention.

0033<figref idref="DRAWINGS">FIG. 20</figref> depicts a flowchart of a method for processing first and second speech signals to produce an output speech signal for playback in accordance with an embodiment of the present invention.

0034<figref idref="DRAWINGS">FIG. 21</figref> depicts a flowchart of a method for updating an amount of gain to be applied to a first speech signal received for playback by an audio device in accordance with an embodiment of the present invention.

0035<figref idref="DRAWINGS">FIG. 22</figref> depicts a waveform plot of an exemplary far-end speech signal that may be processed by an SIE system in accordance with an embodiment of the present invention.

0036<figref idref="DRAWINGS">FIG. 23</figref> depicts a waveform plot of a first output speech signal produced by an SIE system in accordance with an embodiment of the present invention.

0037<figref idref="DRAWINGS">FIG. 24</figref> depicts a waveform plot of a second output speech signal produced by an SIE system in accordance with an embodiment of the present invention.

0038<figref idref="DRAWINGS">FIG. 25</figref> depicts a waveform plot of a third output speech signal produced by an SIE system in accordance with an embodiment of the present invention.

0039<figref idref="DRAWINGS">FIG. 26</figref> is a waveform plot of an exemplary far-end speech signal that may be processed by an SIE system in accordance with an embodiment of the present invention.

0040<figref idref="DRAWINGS">FIG. 27</figref> is a waveform plot of exemplary ambient background noise present in an environment in which a telephony device that includes an SIE system in accordance with an embodiment of the present invention is being used.

0041<figref idref="DRAWINGS">FIG. 28</figref> is a waveform plot of an output speech signal produced by an SIE system in accordance with an embodiment of the present invention responsive to processing the far-end speech signal depicted in the waveform plot of <figref idref="DRAWINGS">FIG. 13</figref> and the near-end background noise depicted in the waveform plot of <figref idref="DRAWINGS">FIG. 14</figref>.

0042<figref idref="DRAWINGS">FIG. 29</figref> depicts waveform plots of audio content presented to the right and left ear of a user to simulate and illustrate the effect of a telephony device that does not include an SIE system in accordance with an embodiment of the present invention.

0043<figref idref="DRAWINGS">FIG. 30</figref> depicts waveform plots of audio content presented to the right and left ear of a user to simulate and illustrate the effect of a telephony device that includes an SIE system in accordance with an embodiment of the present invention.

0044<figref idref="DRAWINGS">FIG. 31</figref> is a block diagram of an example computer system that may be configured to implement an embodiment of the present invention.

0045The features and advantages of the present invention will become more apparent from the detailed description set forth below when taken in conjunction with the drawings, in which like reference characters identify corresponding elements throughout. In the drawings, like reference numbers generally indicate identical, functionally similar, and/or structurally similar elements. The drawing in which an element first appears is indicated by the leftmost digit(s) in the corresponding reference number.

DETAILED DESCRIPTION OF THE INVENTION

A. Introduction

0046The following detailed description refers to the accompanying drawings that illustrate exemplary embodiments of the present invention. However, the scope of the present invention is not limited to these embodiments, but is instead defined by the appended claims. Thus, embodiments beyond those shown in the accompanying drawings, such as modified versions of the illustrated embodiments, may nevertheless be encompassed by the present invention.

0047References in the specification to “one embodiment,” “an embodiment,” “an example embodiment,” or the like, indicate that the embodiment described may include a particular feature, structure, or characteristic, but every embodiment may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Furthermore, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of persons skilled in the relevant art(s) to implement such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.

0048A speech intelligibility enhancement (SIE) system and method is described herein that can be used to improve the intelligibility of a speech signal received over a communication network from a far-end telephony terminal for playback at a near-end telephony terminal. The SIE system and method is particularly useful in a scenario in which a user of the near-end telephony terminal attempts to conduct a telephone call in an environment with loud acoustic background noise, as described in the Background Section above. Generally speaking, the SIE system and method, which may be implemented as part of the near-end telephony terminal, monitors both the speech signal received from the far-end telephony terminal and a near-end background noise signal and, based on both signals, modifies the speech signal to increase the intelligibility while minimizing the distortion thereof.

0049In one embodiment, the SIE system and method increases intelligibility by maintaining a desired minimum signal-to-noise ratio (SNR) between the speech signal being played back on a loudspeaker of the near-end telephony terminal and the ambient background noise. The minimum SNR is determined such that the speech remains intelligible in the presence of the ambient background noise.

0050In a further embodiment, the SIE system and method is configured to attain the minimum SNR by applying a pure linear gain to the speech signal received from the far-end telephony terminal. However, should digital saturation of the output waveform be reached before the minimum SNR has been reached, then the SIE system and method performs amplitude compression to allow greater subsequent amplification of lower level segments of the received speech signal.

0051In accordance with a particular implementation of the SIE system and method, the performance of amplitude compression followed by amplification is carried out in such a manner that digital saturation is impossible. Thus, the system and method is guaranteed never to saturate and cause clipping of the speech output signal. As will be described in more detail herein, this is achieved in part by using a frame-by-frame instant attack approach to tracking the waveform envelope of the received speech signal and then using information derived from such waveform envelope tracking to limit the amount of gain that may ultimately be applied to the received speech signal.

0052In a still further embodiment, the SIE system and method monitors the degree of amplitude compression and uses this information as an input (in a feedback manner) to control an amount of spectral shaping that is applied to the received speech signal. If no amplitude compression is applied, then no spectral shaping is applied since the minimum SNR was attained without amplitude compression. However, if amplitude compression is applied, then this indicates that there was not enough digital “headroom” to attain the minimum SNR through the application of a simple linear gain. The application of amplitude compression allows for further increases in loudness, but at some point this technique is also not sufficient to ensure intelligibility. Hence, at a certain point spectral shaping is enabled based on the amount of amplitude compression. The amount of amplitude compression can also be thought of as the amount of linear gain that had to be discarded (on the highest amplitudes), but was required to reach the minimum SNR. Viewed this way, the amount of amplitude compression provides a good indicator of the remaining shortage of intelligibility, which must be provided using different means in order to reach the minimum SNR for intelligibility. Hence, in accordance with this embodiment, the amount of spectral shaping applied is a function of at least the amount of amplitude compression that was applied.

0053In an alternative embodiment, the amount of spectral shaping applied is a function of the amount of digital headroom (or lack thereof) between the signal level required to achieve the minimum SNR and the digital saturation point or some other point at which amplitude compression will be applied. Note that in additional embodiments, spectral shaping may be applied to the received speech signal in a manner that is not dependent on feedback concerning the degree of amplitude compression or the amount of digital headroom available.

0054The example SIE system and methods described herein may advantageously be implemented in a wide variety of telephony terminals including but not limited to, corded telephones, cordless telephones, cellular telephones, Bluetooth® headsets, or any other telephony terminals configured to pick up and transmit speech signals representative of the voice of a near-end user to a far-end user and to receive and play back speech signals representative of the voice of the far-end user to the near-end user.

B. Speech Intelligibility Enhancement System in Accordance with an Embodiment of the Present Invention

0055<figref idref="DRAWINGS">FIG. 2</figref> is a block diagram of an example SIE system <b>200</b> in accordance with one embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 2</figref>, SIE system <b>200</b> includes a spectral shaping block <b>202</b>, a dispersion filter <b>204</b>, a level estimator <b>206</b>, a waveform envelope tracker <b>208</b>, a compression tracker <b>210</b>, and an Automatic Gain Control (AGC)/Automatic Volume Boosting (AVB)/compression block <b>212</b>. AGC/AVB/compression block <b>212</b> includes AGC logic <b>222</b>, AVB logic <b>224</b>, and compression logic <b>226</b>.

0056Generally speaking, the components of example SIE system <b>200</b> operate together to improve the intelligibility of a speech signal received over a communication network from a far-end telephony terminal (referred to herein as the “far-end speech signal”) for playback by a near-end telephony terminal of which SIE system <b>200</b> is a part. In <figref idref="DRAWINGS">FIG. 2</figref>, the far-end speech signal is denoted “Receive-in.” This signal may be received from another component in the telephony terminal. For example, the far-end speech signal may be received from a buffer that stores digital samples produced by an audio decoder within the telephony terminal. The audio decoder in turn may produce the digital samples by decoding an encoded bit stream transported over a communication network. The output of SIE system <b>200</b> is the modified far-end speech signal, denoted “Receive-out” in <figref idref="DRAWINGS">FIG. 2</figref>, which is provided directly or indirectly to a loudspeaker for playback to a user.

0057Certain components of system <b>200</b> will now be briefly described and additional details about each component will be provided in the sub-sections below.

0058AGC logic <b>222</b> is configured to compensate for variations in the level of the far-end speech signal. For example, such variations may be due to variation of network connections, acoustic coupling, or the like. AGC logic <b>222</b> calculates a gain scaling factor that, when applied to the far-end speech signal, brings the far-end speech signal to a nominal signal level.

0059AVB logic <b>224</b> is configured to automatically boost the level of the far-end speech signal to maintain at least a minimum SNR as the level of near-end background noise increases. In particular, AVB logic <b>224</b> is configured to maintain at least a predetermined minimum far-end speech signal to near-end noise ratio by calculating an additional gain to be applied to the far-end speech signal if the level of the near-end background noise is such that the level of the far-end speech signal after AGC yields an SNR below the predetermined minimum SNR.

0060Level estimator <b>206</b> is configured to determine an estimated level of the far-end speech signal and to provide this information to AGC logic <b>222</b> and AVB logic <b>224</b> for use in performing gain calculations.

0061Compression logic <b>226</b> is configured to apply a time-varying gain to the far-end speech signal that allows for application of the full AVB gain to attain the desired minimum SNR without digital saturation or clipping of the output signal. In determining the time varying gain, compression logic <b>226</b> takes into account all the gains to be applied to the far-end speech signal before playback (for example, user volume gain, echo suppression gain, or the like). In one implementation, a single gain is applied to the far-end speech signal to achieve the intended effect while in an alternate implementation a separate gain is applied by each of AGC logic <b>222</b>, AVB logic <b>224</b> and compression logic <b>226</b> in order to achieve the intended effect.

0062Generally speaking, compression logic <b>226</b> operates by applying more attenuation to larger waveform peaks than to lower peaks. Effectively, compression logic <b>226</b> boosts the low-amplitude regions of the far-end speech signal when AVB logic <b>224</b> cannot maintain the intelligibility of the far-end speech signal without causing saturation. In particular, compression logic <b>226</b> applies smaller gains to the high-amplitude regions of the far-end speech signal and larger gains to the low-amplitude regions. This has the effect of compressing the high-amplitude regions relative to the low-amplitude regions, thus the name. Such amplitude compression may be simply referred to as “compression” elsewhere in this document as shorthand. In effect, compression logic <b>226</b> amplifies the low-amplitude regions relative to the high-amplitude regions without exceeding the digital saturation level, and therefore has the effect of increasing the loudness of the far-end speech signal without introducing digital saturation.

0063Waveform envelope tracker <b>208</b> is configured to perform waveform envelope tracking on the far-end speech signal and to provide waveform tracking information to AGC/AVB/compressor block <b>212</b> that can be used by that block to determine exactly how much headroom there is to digital saturation in the far-end speech signal prior to modifying it.

0064Dispersion filter <b>204</b> is configured to reduce a peak-to-average ratio of the waveform samples of the far-end speech signal so that the filtered speech signal has smaller peak values and thus allows more headroom for AVB logic <b>224</b> to boost the far-end speech signal without introducing digital saturation. In an exemplary embodiment of the present invention, dispersion filtering is achieved using an all-pass filter. Such an all-pass filter can be either fixed or adaptive. A fixed all-pass filter is lower in complexity but can achieve only a smaller reduction of the magnitude peak of the far-end speech. Conversely, an adaptive all-pass filter has higher complexity but also has the potential to achieve a larger reduction of the magnitude peak.

0065Spectral shaping block <b>202</b> is configured to boost certain local peaks of the spectral envelope (called “formants”) of the far-end speech signal above the near-end noise floor to make the far-end speech signal more intelligible. In particular, spectral shaping block <b>202</b> is configured to boost certain formants of the far-end speech signal above the spectral values of the near-end noise at corresponding frequencies. In trying to understand spoken speech, humans normally rely on recognizing the frequencies of the speech formants. Therefore, by boosting certain formants of the far-end speech signal above the noise floor, spectral shaping block <b>202</b> makes the far-end speech more intelligible. In one embodiment, the second and third formants of the far-end speech signal are boosted relative to the first formant since the second and third formants are more important from the perspective of speech intelligibility than the first formant.

0066In one exemplary embodiment of the present invention, spectral shaping is implemented by adaptive high-pass filtering. For example, such adaptive high-pass filtering may be used to boost the second and third formants of the far-end speech signal relative to the first formant, since the second and third formants are located at higher frequencies than the first formant. The degree of high-pass filtering may depend on the far-end speech as well as the near-end noise. The high-pass filter may consist of a single-stage filter or multiple stages of filters, where different stages have different adaptation characteristics. For example, the high-pass filter may contain two stages of high-pass filters, with a slowly-evolving first stage having a long adaptation time constant and a rapidly-evolving second stage having a relatively short adaptation time constant.

0067In accordance with one implementation of SIE system <b>200</b>, the signal processing techniques performed by AGC logic <b>222</b>, AVB logic <b>224</b>, compression logic <b>226</b>, dispersion filter <b>204</b> and spectral shaping block <b>202</b> are applied one-by-one in a specific sequence so as to minimize the distortion introduced to the far-end speech signal, with each new technique being applied only when necessary. For example, AGC may first be applied by AGC logic <b>222</b> to bring the far-end speech to a nominal level. If the background noise level is low, AGC may be the only technique applied. As the background noise level increases, AVB may be applied by AVB logic <b>224</b> to increase the volume of the far-end speech signal. As the background noise level increases further, compression may then be applied by compression logic <b>226</b> to further boost the low-amplitude regions of the far-end speech signal if AVB is not sufficient to maintain the intelligibility of the far-end speech signal. As the background noise level increases even further, dispersion filtering can be applied by dispersion filter <b>204</b> to reduce the peak-to-average ratio of the far-end speech signal, thereby providing additional headroom for performing AVB. If the background noise is so loud that the above four techniques are not sufficient, spectral shaping can then be applied by spectral shaping block <b>202</b> to further enhance the speech intelligibility by exploiting the properties of human perception.

0068With further reference to the foregoing example implementation, AGC and AVB are applied first since those techniques hardly introduce any distortion to the far-end speech signal. Compression however can make speech sound slightly unnatural due to the compression of natural dynamic range, and dispersion filtering may introduce a slight distortion to the speech; therefore, these two techniques are applied only when AGC and AVB alone cannot provide sufficient intelligibility of the far-end speech signal. Finally, depending on the telephony terminal, spectral shaping may make the most dramatic and audible modification of the far-end speech signal and thus this technique is only applied when the above four techniques do not provide sufficient intelligibility of the far-end speech.

0069In alternate implementations, exceptions to this approach may be made. For example, in certain embodiments techniques that increase distortion in a traditional sense are applied before the amount of linear gain that may be applied without reaching digital saturation has been exhausted. One example of such an embodiment is an embodiment that limits high waveform amplitudes below a maximum digital amplitude to protect the auditory system of a user from exposure to uncomfortable, or possibly, damaging signal levels.

0070Each of the foregoing components of system <b>200</b> and the manner in which such components operate to implement aspects of the present invention will now be described. In the following description, it is assumed that the speech signal being processed comprises a series of digital samples and that the series of digital samples is divided into discrete time segments termed frames. In the description, individual frames are referred to by a frame counter, wherein a frame counter k generally refers to the frame currently being processed and frame counter k−1 refers to the immediately previous frame.

0071It should be understood that while most of the algorithm parameters given below are specified assuming a sampling rate of 8 kHz for telephone-bandwidth speech, persons skilled in the relevant art(s) should have no problem extending the techniques presented below to other sampling rates, such as 16 kHz for wideband speech. Therefore, the parameters specified are only meant to be exemplary values and are not limiting.

00721. Spectral Shaping Block <b>202</b>

0073In SIE system <b>200</b>, spectral shaping block <b>202</b> is configured to receive the far-end speech signal (shown as “Receive-in” in <figref idref="DRAWINGS">FIG. 2</figref>) and to apply spectral shaping thereto in a manner that is controlled by feedback from compression tracker <b>210</b>. As will be described in more detail below, such spectral shaping may include both slowly-evolving and rapidly-evolving spectral shaping filters, wherein the combination offers the advantage of not having to drive either filter too hard.

0074Spectral shaping block <b>202</b> is configured to boost certain formants of the far-end speech signal above the near-end noise floor so that they can be recognized by the near-end telephony terminal user and thus help that user understand the speech. Since the far-end speech signal is changing with time, such spectral shaping is preferably adaptive in order to increase effectiveness. Also, to avoid introducing distortion, such spectral shaping is preferably evolved in a smooth manner.

0075One possible manner of performing such spectral shaping is to perform spectral analysis followed by synthesis. This may be accomplished by using a Fast Fourier Transform (FFT) and inverse FFT, or using sub-band analysis and sub-band synthesis. For example, with FFT or sub-band analysis of both the far-end speech and the near-end noise, a determination can be made as to whether the formants of the far-end speech signal are below the noise floor. If so, those spectral components of the far end speech signal around the formants are boosted (i.e., a gain is applied) such that they are at least Y dB above the noise floor, where Y is determined and tuned empirically. Then, the modified frequency-domain representation of the far-end speech is converted back to a time domain signal.

0076Although the foregoing method allows for precise control of the SNR at each formant frequency, one drawback of the method is that it requires significant complexity. In an exemplary embodiment of the present invention, the spectral shaping is achieved with very-low-complexity time-domain filtering using a low-order high-pass filter. The use of such a high-pass filter achieves two goals. First, it helps to boost the second and third formants of the far-end speech signal. The second and third formants are more critical to speech intelligibility and are often much lower in intensity as compared with the first formant and thus are frequently buried under the noise floor when in a noisy environment. Second, it attenuates the first formant around or below 500 Hz, which normally dominates the energy content of the voiced speech signal and which often overloads the tiny loudspeakers used in many telephony devices. By attenuating the first formant relative to the second and third formants, the high-pass filter allows more energy that is useful for intelligibility to be emitted from such tiny loudspeakers before overloading them.

0077In one embodiment of the present invention, the high-pass spectral shaping filter consists of two cascaded filters: a slowly-evolving spectral shaping filter and a rapidly-evolving spectral shaping filter, each of which is controlled by different adaptation mechanisms. <figref idref="DRAWINGS">FIG. 3</figref> depicts a block diagram of such a high-pass spectral shaping filter <b>300</b>. As shown in <figref idref="DRAWINGS">FIG. 3</figref>, the high-pass spectral shaping filter <b>300</b> consists of a slowly-evolving spectral shaping filter <b>302</b> and a rapidly-evolving spectral shaping filter <b>304</b>.

0078In accordance with one implementation, slowly-evolving spectral shaping filter <b>302</b> has the form of <br /><i>x</i>(<i>n</i>)=<i>r</i>_in(<i>n</i>)−<i>b·r</i>_in(<i>n</i>−1),  (1)<br /> where x(n) is the output, r_in(n) is the input, and b is the filter coefficient. The filter coefficient is determined according to a table lookup <br /><i>b=b</i>_tbl<i>[idx],</i>  (2)<br /> where the table can be <br /><i>b</i>_tbl[ ]={0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8},  (3)<br /> and the index is determined according to

0079<maths id="MATH-US-00001" num="00001"><math overflow="scroll"><mtable><mtr><mtd><mrow><mi>idx</mi><mo>=</mo><mrow><mrow><mo>(</mo><mrow><msub><mi>N</mi><msub><mi>b</mi><mi>tbl</mi></msub></msub><mo>-</mo><mn>1</mn></mrow><mo>)</mo></mrow><mo>·</mo><mrow><mo>⌊</mo><mfrac><mrow><mi>min</mi><mo>⁡</mo><mrow><mo>(</mo><mrow><mrow><msub><mi>V</mi><mi>loss</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>-</mo><mn>1</mn></mrow><mo>)</mo></mrow></mrow><mo>,</mo><msub><mi>mxV</mi><mi>loss</mi></msub></mrow><mo>)</mo></mrow></mrow><msub><mi>mxV</mi><mi>loss</mi></msub></mfrac><mo>⌋</mo></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>4</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0001.tif" /><br /> in which N_b_{<sub2>tbl </sub2>}is the table size, e.g., N_b_{<sub2>tbl</sub2>}=9 above, V_loss(k−1) the smoothed volume loss (or loss in headroom) due to compression applied by compression logic <b>226</b> as tracked by compression tracker <b>210</b>, and mxV_lossis a smoothed volume loss at which maximum slowly varying spectral shaping is applied, e.g., mxV_loss=27.

0080The frequency response of the filters given by the coefficients in Eq. 3 and the filter of Eq. 1 are shown in graph <b>400</b> of <figref idref="DRAWINGS">FIG. 4</figref>. As can be seen, the filters will generally attenuate the first formant while amplifying formants <b>2</b> and above, thereby increasing intelligibility. In a possible configuration for wideband speech where this filter is applied to the 0-4 kHz band, a constant gain can be applied to the 4-8 kHz band to prevent a spectral discontinuity at 4 kHz, and instead facilitate a continuous full-band modification of the signal. The gain for the 4-8 kHz band would depend on the filter coefficient. The gains corresponding to the filter coefficients of Eq. 3 are {1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8}.

0081In one implementation, rapidly-evolving spectral shaping filter <b>304</b> includes two control parameters. The first control parameter is given by <br />ratio=1−10^−(αV^{<sub2>loss</sub2>}^(k−1)/20),  (5)<br /> where α is a control parameter, e.g. α=0.375. The second control parameter is given by

0082<maths id="MATH-US-00002" num="00002"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><mi>rho</mi><mo>=</mo><mfrac><mrow><msub><mi>r</mi><mi>sm</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>,</mo><mn>1</mn></mrow><mo>)</mo></mrow></mrow><mrow><msub><mi>r</mi><mi>sm</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>,</mo><mn>0</mn></mrow><mo>)</mo></mrow></mrow></mfrac></mrow><mo>,</mo><mstyle><mtext></mtext></mstyle><mo>⁢</mo><mi>where</mi></mrow></mtd><mtd><mrow><mo>(</mo><mn>6</mn><mo>)</mo></mrow></mtd></mtr><mtr><mtd><mrow><mrow><msub><mi>r</mi><mi>sm</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>,</mo><mi>m</mi></mrow><mo>)</mo></mrow></mrow><mo>=</mo><mrow><mrow><mi>λ</mi><mo>·</mo><mrow><msub><mi>r</mi><mi>sm</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mrow><mi>k</mi><mo>-</mo><mn>1</mn></mrow><mo>,</mo><mi>m</mi></mrow><mo>)</mo></mrow></mrow></mrow><mo>+</mo><mrow><mrow><mo>(</mo><mrow><mn>1</mn><mo>-</mo><mi>λ</mi></mrow><mo>)</mo></mrow><mo>·</mo><mrow><msub><mi>r</mi><mi>cor</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mi>m</mi><mo>)</mo></mrow></mrow></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>7</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0002.tif" /><br /> The smoothing constant λ can have a value of 0.75, for example. In the equation the auto correlation is calculated as

0083<maths id="MATH-US-00003" num="00003"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><msub><mi>r</mi><mi>cor</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mi>m</mi><mo>)</mo></mrow></mrow><mo>=</mo><mrow><munderover><mo>∑</mo><mrow><mi>n</mi><mo>=</mo><mn>0</mn></mrow><mrow><mi>N</mi><mo>-</mo><mi>m</mi></mrow></munderover><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><mrow><msub><mi>r</mi><mrow><mi>i</mi><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mi>n</mi></mrow></msub><mo>⁡</mo><mrow><mo>(</mo><mi>n</mi><mo>)</mo></mrow></mrow><mo>·</mo><mrow><msub><mi>r</mi><mrow><mi>i</mi><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mi>n</mi></mrow></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>n</mi><mo>-</mo><mi>m</mi></mrow><mo>)</mo></mrow></mrow></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>8</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0003.tif" /><br /> where N is the frame size, e.g. 40 samples, corresponding to 5 ms at 8 kHz. The final filter coefficient of rapidly evolving spectral shaping filter <b>304</b> is given by <br /><i>c</i>=max(γ·rho·ratio,0),  (9)<br /> where γ controls the maximum filter coefficient, e.g. γ=0.75. The filter equation for the rapidly evolving spectral shaping is given by <br /><i>y</i>(<i>n</i>)=<i>x</i>(<i>n</i>)−<i>c·x</i>(<i>n−</i>2)−<i>c·y</i>(<i>n−</i>1).  (10)

0084In accordance with the foregoing, rapidly-evolving spectral shaping filter <b>304</b> is a second-order pole-zero high-pass filter having one pole and two zeros, with a transfer function of

0085<maths id="MATH-US-00004" num="00004"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><msub><mi>H</mi><mi>re</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mi>z</mi><mo>)</mo></mrow></mrow><mo>=</mo><mfrac><mrow><mn>1</mn><mo>-</mo><msup><mi>cz</mi><mrow><mo>-</mo><mn>2</mn></mrow></msup></mrow><mrow><mn>1</mn><mo>+</mo><msup><mi>cz</mi><mrow><mo>-</mo><mn>1</mn></mrow></msup></mrow></mfrac></mrow></mtd><mtd><mrow><mo>(</mo><mn>11</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0004.tif" /><br /> where c is the single parameter that controls the shape of the frequency response of the filter. The family of frequency response curves for different values of c is plotted in graph <b>500</b> of <figref idref="DRAWINGS">FIG. 5</figref>. This filter is designed to be totally controlled by a single parameter c. This makes it simple to implement and to adapt from frame to frame.

0086Rapidly-evolving spectral shaping filter <b>304</b> is designed to have relatively sharp attenuation at or below about 500 Hz, where the first formant of voiced speech usually is. Also, it boosts the second and third formants relative to the first formant. This filter is also designed to have exactly the same magnitude response value of 0 dB at half the sampling frequency. This makes it easier to achieve a seamless transition to a higher band when using a split-band system in wideband applications. In other words, a high band filter can always start at 0 dB no matter what the value of the filter control parameter c is, and the corresponding composite magnitude response will always be continuous at the band boundary of the low band (where this filter is) and the high band.

0087Another important feature is that at frequencies above 3400 Hz, the magnitude responses in <figref idref="DRAWINGS">FIG. 5</figref> always go down toward 0 dB. This arrangement has the desirable effect of not excessively amplifying the potential noise in the far-end speech signal in the stop band of 3400 to 4000 Hz.

0088Slowly-evolving spectral shaping filter <b>302</b> and rapidly-evolving spectral shaping filter <b>304</b> can be combined into a single spectral shaping filter, if desired, by convolving the filter response of slowly evolving spectral shaping filter <b>302</b> with the zero section of rapidly evolving spectral shaping filter <b>304</b>, and maintaining the pole section of rapidly evolving spectral shaping filter <b>304</b>.

0089Note that in the specific implementation discussed above, the operation of slowly-evolving spectral shaping filter <b>302</b> and rapidly-evolving spectral shaping filter <b>304</b> is controlled, in part, by V_loss(k−1), which is the smoothed volume loss (or loss in headroom) resulting from compression applied by compression logic <b>226</b> and fed back by compression tracker <b>210</b>. The smoothed volume loss provides an indication of the remaining shortage of intelligibility in the far-end speech signal after the application of compression thereto. This shortage must be compensated for using different means in order to reach a minimum SNR for intelligibility. Hence, in accordance with this embodiment, the amount of spectral shaping applied is a function of the smoothed volume loss.

0090However, the present invention is not limited to this approach and spectral shaping may be applied to the far-end speech signal in a manner that is not controlled by the smoothed volume loss or by any other measurement of the degree of compression applied to the far-end speech signal by compression logic <b>226</b>.

0091Furthermore, although spectral shaping is described herein as one of a plurality of techniques used for performing SIE, persons skilled in the relevant art(s) will appreciate that spectral shaping alone can be used to enhance speech intelligibility.

00922. Dispersion Filter <b>204</b>

0093As shown in <figref idref="DRAWINGS">FIG. 2</figref>, dispersion filter <b>204</b> may be inserted after spectral shaping block <b>202</b> but before an input level estimator <b>206</b>. Depending upon the implementation, dispersion filter <b>204</b> could also be merged with the spectral shaping filter(s) in spectral shaping block <b>202</b> to form a single filter, or it could be moved ahead of spectral shaping block <b>202</b>.

0094The function of dispersion filter <b>204</b> is to reduce the peak-to-average ratio for waveform sample magnitudes of the far-end speech signal. One way to measure the “average” is the Root-Mean-Square (RMS) value that is well-known in the art. Some of the speech vowel signals are fairly “peaky”—that is, they have a high peak-to-RMS ratio. In this case, such speech signals cannot be amplified to a very loud level before the waveform peaks are clipped at digital saturation level. Dispersion filter <b>204</b> can “disperse” or effectively smear out such waveform peaks so that the energy of the waveform peak is more evenly distributed across the time axis after such filtering. When it achieves this, the peak-to-RMS ratio is reduced. In other words, for the same RMS value or energy level, the waveform magnitude peak is reduced, leaving more “headroom” to digital saturation for AVB logic <b>224</b> to utilize. The waveform can then be amplified more before clipping occurs, and this will boost the effective volume of the far-end speech signal and enhance the speech intelligibility. Generally speaking, if dispersion filter <b>204</b> can reduce the peak-to-RMS ratio by X dB and if AVB logic <b>224</b> can fully utilize this additional X dB of headroom, then after application of AVB the output signal level will be X dB higher without clipping.

0095There are many ways to perform dispersion filtering. Since one of the objectives of the SIE system and method is to minimize the distortion introduced to the far-end speech signal, an exemplary embodiment of the present invention uses an all-pass filter as the dispersion filter, because an all-pass filter has a completely flat magnitude frequency response of value 1 and thus does not introduce any magnitude distortion whatsoever. The only distortion it can introduce is phase distortion, but human ears are generally not very sensitive to phase distortion.

0096Since the magnitude frequency response of an all-pass filter has value 1 for all frequencies, the input signal and the output signal of an all-pass filter have exactly the same RMS value. Therefore, with all-pass filtering, minimizing the peak-to-RMS ratio is exactly the same as minimizing the waveform peak value.

0097As is well-known in the art, an all-pass filter is a pole-zero filter with the numerator polynomial and denominator polynomial of its transfer function sharing the same set of polynomial coefficients except with the order reversed. With proper design, even a fixed 6^th-order all-pass filter can provide on average nearly 2 dB of reduction in the peak-to-RMS ratio of high-magnitude speech vowel signals. An example transfer function of such a fixed 6^th-order all-pass filter optimized for 8 kHz sampled speech is given below.

0098<maths id="MATH-US-00005" num="00005"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><mi>H</mi><mo>⁡</mo><mrow><mo>(</mo><mi>z</mi><mo>)</mo></mrow></mrow><mo>=</mo><mfrac><mrow><msub><mi>a</mi><mn>6</mn></msub><mo>+</mo><mrow><msub><mi>a</mi><mn>5</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>1</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>4</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>2</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>3</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>3</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>2</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>4</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>1</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>5</mn></mrow></msup></mrow><mo>+</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>6</mn></mrow></msup></mrow><mrow><mn>1</mn><mo>+</mo><mrow><msub><mi>a</mi><mn>1</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>1</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>2</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>2</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>3</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>3</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>4</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>4</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>5</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>5</mn></mrow></msup></mrow><mo>+</mo><mrow><msub><mi>a</mi><mn>6</mn></msub><mo>⁢</mo><msup><mi>z</mi><mrow><mo>-</mo><mn>6</mn></mrow></msup></mrow></mrow></mfrac></mrow></mtd><mtd><mrow><mo>(</mo><mn>12</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0005.tif" /><br /> The filter coefficients may be, for example, a₁=−1.787, a₂=2.432, a₃=−2.565, a₄=2.171, a₅=−1.408, a₆=0.699. An exemplary embodiment of the present invention can use such a fixed all-pass filter. Using such a fixed all-pass filter has the advantage of relatively low complexity.

0099It is also possible to make the all-pass filter adaptive to achieve more waveform peak reduction, albeit at the cost of higher complexity. The poles and zeros of the all-pass filter can be adapted according to the local characteristics of the speech waveform so as to maximize the reduction of the waveform peak magnitude. In addition, just as the smoothed volume loss, V_loss(k), can be used to control the spectral shaping filter(s) in spectral shaping block <b>202</b>, V_loss(k) can also be used to control an adaptive all-pass filter. For example, similarly to how the spectral shaping is gradually increased by an increasing V_loss(k), the amount of dispersion can be gradually increased by an increasing V_loss(k). This can be achieved by mapping the V_loss(k) to a scaling factor that is applied to the radii of the poles of the dispersion filter. The mapping maps a low V_loss(k) to a scaling factor close to zero (effectively disabling dispersion), and a high V_loss(k) to a scaling factor close to one (allowing full dispersion). The usage of V_loss(k) to control the dispersion is shown by the dashed line connecting compression tracker <b>210</b> to dispersion filter <b>204</b> in <figref idref="DRAWINGS">FIG. 2</figref>.

0100The effect of all-pass dispersion filtering is illustrated in <figref idref="DRAWINGS">FIG. 6</figref>, where an example male speech waveform before the dispersion filtering is shown in an upper plot <b>602</b>, and the same segment of speech waveform after dispersion filtering is shown in a lower plot <b>604</b>. The two horizontal dashed lines represent the lines corresponding to zero signal magnitude for these two waveforms, respectively. Note that the two waveforms have identical energy values and even sound essentially the same, because the dispersion filter used was an all-pass filter.

0101It can be seen from <figref idref="DRAWINGS">FIG. 6</figref> that the waveform in upper plot <b>602</b> has about five periods of nearly periodic pitch cycle waveform, where each period has a sharp negative peak. After all-pass dispersion filtering, these sharp negative peaks were spread out into many smaller peaks, and the maximum signal magnitude is reduced in the process. Specifically, the speech waveform in upper plot <b>602</b> has the largest negative peak in the middle of the plot with a magnitude of <b>8822</b> in a 16-bit linear PCM representation. After all-pass dispersion filtering, the filter output signal in lower plot <b>604</b> has a maximum magnitude of <b>4544</b>. This represents a peak magnitude reduction of 5.76 dB. In the ideal situation in which AVB logic <b>224</b> can fully utilize this reduced peak magnitude (i.e. increased “digital headroom”), AVB logic <b>224</b> can boost the intensity of the signal in the lower plot 5.76 dB more than it can boost the intensity of the signal in the upper plot before reaching the digital saturation level. Therefore, in this example of <figref idref="DRAWINGS">FIG. 6</figref>, compared with the unfiltered signal shown in plot <b>602</b>, the signal after dispersion filtering shown in plot <b>604</b> can be boosted to be 5.76 dB higher in intensity in an ideal situation.

0102A similar waveform plot for an example female speech signal is shown in <figref idref="DRAWINGS">FIG. 7</figref>. In particular, an example female speech waveform before dispersion filtering is shown in an upper plot <b>702</b>, and the same segment of speech waveform after dispersion filtering is shown in a lower plot <b>704</b>. In this case, the sharp positive waveform peaks in upper plot <b>702</b> were reduced in lower plot <b>704</b>, and the all-pass filter reduced the peak magnitude by 4.44 dB. In both <figref idref="DRAWINGS">FIG. 6</figref> and <figref idref="DRAWINGS">FIG. 7</figref>, a 6^th-order all-pass filter optimized for that segment of speech signal was used.

0103Through experiments, it was found that the optimal all-pass filter for a given frame of voiced speech signal usually has its poles and zeros located near but not exactly at the speech formant frequencies. (Here “optimal” is in the sense of minimizing the peak-to-RMS ratio, or equivalently, minimizing the waveform peak magnitude.) Also, it was found that the degree of waveform peak reduction is controlled by how close the poles (and the corresponding zeros) of the all-pass filter are to the unit cycle. As the radii of the poles approach the range of 0.90 to 0.95, large waveform peak reduction can be achieved. On the other hand, as the radii of the poles approaches zero, the effect of all-pass filtering gradually diminishes.

0104Based on such an observation, an exemplary embodiment of the present invention employs an adaptive all-pass filter where the radii of its poles are set at or near zero during silence regions of the far-end speech signal and are adapted toward the range of 0.90 to 0.95 during high-magnitude vowel regions. Also, at or near the beginning of a voiced region of the far-end speech signal, the frequencies (or equivalently, polar angles) of the poles of the adaptive all-pass filter are set to the pole frequencies of an optimal fixed all-pass filter, such as the 6^th-order fixed all-pass filter shown above. Then, during the syllable of that vowel sound, the pole frequencies are adapted frame by frame to try to maintain near optimality by tracking the change in that vowel speech signal. One example way of performing such tracking is to estimate the formant frequencies and then use such formant frequencies to guide the adaptation of the pole frequencies of the all-pass filter (with the corresponding changes to the frequencies of the zeros).

0105The estimate of formant frequencies need not be very accurate, and certainly not to the same degree of accuracy required by some formant-based speech synthesis systems. Basically, in terms of minimizing the waveform magnitude peak, what matters is the relative phase relationship between pitch harmonics near prominent peaks of the spectral envelope of the speech signal. Therefore, even a crude estimation of rough formant frequencies based on picking frequencies of spectral peaks in the frequency response of a short-term predictive synthesis filter (often called the “LPC filter” in speech coding literature) will suffice.

0106In addition to (or in place of) such guidance from estimated formant frequencies, one can also use a closed-loop pole frequency search to find the optimal pole frequencies and to guide the adaptation of such pole frequencies. It was found that when an all-pass filter is used, the pole frequencies cannot change too much from frame to frame, otherwise there will be a significant difference in the group delays of the filtered signals in the adjacent frames which will cause an audible waveform disturbance. To minimize the possibility of such distortion, the closed-loop pole frequency search limits this search range to be in the neighborhoods of the previous pole frequencies. It was found that a frequency resolution of 5 to 10 Hz is sufficient to achieve most of the magnitude peak reduction. Therefore, a few pole frequency candidates, which are in the neighborhood of the pole frequencies used in the last frame and which are 5 to 10 Hz away from each other, are tried, and the set of pole frequencies achieving the maximum waveform peak reduction subject to a constraint of tracking the formant trajectory is selected as the winner for the current frame, and the all-pass filter is constructed from this set of pole frequencies and a given set of default pole radii.

0107In the example all-pass filters described above, a filter order of 6 was used because that gives three pole pairs (and the corresponding three zero pairs), which are sufficient to track the first three formants in speech signals that account for most of the speech energy. During the search of the optimal pole frequencies for the adaptive all-pass filter, it is advantageous in terms of computational complexity to search one pole pair at a time. For example, the frequency of the first pole pair can be searched in the frequency range of the first speech formant (typically 270 to 730 Hz) using a frequency grid of 5 to 10 Hz. After the frequency of the first pole pair that minimizes the waveform peak magnitude is identified, with the first pole pair fixed at that optimal frequency and with the effect of the first pole pair taken into account, the frequency of the second pole pair can then be searched in the frequency range of the second speech formant (typically 840 to 2290 Hz). Similarly, after the optimal frequency of the second pole pair is also identified and the effect of the optimal second pole pair taken into account, the frequency of the third pole pair can be searched in the frequency range of the third speech formant (typically 1690 to 3010 Hz). It is also possible to do joint optimization of the frequencies of the pole pairs. Although it has a higher complexity, an adaptive all-pass filter has the potential of achieving significantly more waveform peak reduction than a fixed all-pass filter.

0108Besides a fixed all-pass filter and a fully adaptive all-pass filter, a third possible implementation for dispersion filter <b>204</b> is a switched-adaptive all-pass filter, which achieves a compromise between a fixed all-pass filter and a fully adaptive all-pass filter in terms of complexity and performance. In a switched-adaptive all-pass filter, a collection of N all-pass filter candidates are carefully pre-designed and optimized. Then, in actual filtering, each of the N filter candidates is tried, and the system identifies the filter candidate that minimizes the speech waveform peak magnitude while also satisfying the constraint that the differences between the pole locations (or group delays) of filters in adjacent frames are below pre-set thresholds. Simulations have shown that such a switched-adaptive all-pass filter can achieve significant improvement in waveform peak magnitude reduction over a fixed all-pass filter while also avoiding the waveform distortion due to significant difference between group delays of the filter output signals of adjacent frames.

01093. Level Estimator <b>206</b>

0110In SIE system <b>200</b>, level estimator <b>206</b> is configured to perform level estimation on the signal output from dispersion filter <b>204</b> (i.e., the far-end speech signal after spectral shaping and dispersion filtering have been applied thereto). However, depending upon the implementation, the level of the original far-end speech signal input to spectral shaping block <b>202</b> can instead be estimated, or level estimation can be performed on both the signal input to spectral shaping block <b>202</b> and the signal output from dispersion filter <b>204</b>. However, for complexity considerations it may be desirable to perform level estimation on only one of the signals, and in practice SIE system <b>200</b> will perform satisfactorily when level estimation is performed only on the output of dispersion filter <b>204</b>. As shown in <figref idref="DRAWINGS">FIG. 2</figref>, in one embodiment, another component within the telephony device in which SIE system <b>200</b> is implemented provides a measure of voice activity in the receive-in signal as input to level estimator <b>206</b>. For example, the other component may be a sub-band acoustic echo canceller (SBAEC). The measure of voice activity can be implemented in many ways. One example is to count the number of sub-bands where the energy significantly exceeds the noise floor.

01114. Waveform Envelope Tracker <b>208</b>

0112Waveform envelope tracker <b>208</b> is configured to perform waveform envelope tracking on the signal output from dispersion filter <b>204</b> (i.e., the far-end speech signal after spectral shaping and dispersion filtering have been applied thereto) and to provide waveform tracking information to AGC/AVB/compressor block <b>212</b>. This allows AGC/AVB/compressor block <b>212</b> to determine exactly how much headroom there is to digital saturation in the signal prior to modifying it. In one embodiment, waveform envelope tracker <b>208</b> is configured to calculate the maximum absolute amplitude of the signal waveform in the current frame, e.g. 5 milliseconds (ms). In further accordance with this embodiment, waveform envelope tracker <b>208</b> also maintains a buffer of the maximum absolute amplitudes of the past two 5 ms frames. This allows waveform envelope tracker <b>208</b> to calculate the maximum absolute amplitude of the signal waveform over the past 15 ms. The intent in covering 15 ms is to make sure that at least one pitch period is considered in the maximum. For some talkers of particular low pitch frequency it may be advantageous to increase this value from 15 ms to a larger value. In accordance with this embodiment, waveform envelope tracker <b>208</b> calculates the waveform tracking information as <br /><i>mx</i>(<i>k</i>)=max[φ, 15/16<i>·mx</i>(<i>k−</i>1))],  (13)<br /> where k is the frame counter and φ is the maximum absolute amplitude of the signal waveform over the past 15 ms. Effectively, this embodiment of waveform envelope tracker <b>208</b> provides instant attack and exponential decay.

01135. AGC/AVB/Compressor Block <b>212</b>

0114<figref idref="DRAWINGS">FIG. 8</figref> is a block diagram that depicts AGC/AVB/compression block <b>212</b> of <figref idref="DRAWINGS">FIG. 2</figref> in more detail in accordance with an embodiment of the present invention.

0115The manner in which this particular embodiment of AGC/AVB/compression block <b>212</b> operates will now be described. It is noted that all gain and volume arithmetic described in this section is carried out in the log domain.

0116First, AGC logic <b>222</b>, if enabled, calculates a logarithmic AGC gain to bring the input signal (i.e., the signal output from dispersion filter <b>204</b>) to a predefined nominal level: <br /><i>G</i>_AGC<i>=L</i>_nom<i>−L</i>_R,  (14)<br /> where L_nomis the predefined nominal level and L_Ris the estimated input level as provided by level estimator <b>206</b>. In one embodiment, G_AGCis subject to a minimum and maximum, e.g. −20 dB and +20 dB. An alternate implementation of this AGC logic is described below in Section D.

0117Subsequently, AVB logic <b>224</b> calculates the receive-to-ambient-background-noise ratio after AGC as <br /><i>R</i>2<i>S</i>noise=default_volume+<i>G</i>_AGC<i>+L</i>_R<i>+C−L</i>_Snoise,  (15)<br /> where default_volume is a constant representing a volume providing a comfortable listening level in quiet conditions, L_Snoiseis the estimated ambient noise level, and C is a calibration term to ensure that R2Snoise reflects what the user is experiencing. In one embodiment, the parameter L_Snoisemay be provided from another component within the telephony device in which SIE system <b>200</b> is implemented. For example, the other component may be a sub-band acoustic echo canceller (SBAEC).

0118AVB logic <b>224</b> then calculates the target AVB gain as

0119<maths id="MATH-US-00006" num="00006"><math overflow="scroll"><mtable><mtr><mtd><mrow><msub><mi>TG</mi><mi>AVB</mi></msub><mo>=</mo><mrow><mo>{</mo><mrow><mtable><mtr><mtd><mn>0</mn></mtd><mtd><mrow><mrow><mi>R</mi><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mn>2</mn><mo>⁢</mo><mi>Snoise</mi></mrow><mo>></mo><mrow><mi>TR</mi><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mn>2</mn><mo>⁢</mo><mi>Snoise</mi></mrow></mrow></mtd></mtr><mtr><mtd><mrow><mi>min</mi><mo>⁡</mo><mrow><mo>[</mo><mrow><mrow><mrow><mi>TR</mi><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mn>2</mn><mo>⁢</mo><mi>Snoise</mi></mrow><mo>-</mo><mrow><mi>R</mi><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mn>2</mn><mo>⁢</mo><mi>Snoise</mi></mrow></mrow><mo>,</mo><msub><mi>mxG</mi><mi>AVB</mi></msub></mrow><mo>]</mo></mrow></mrow></mtd><mtd><mi>otherwise</mi></mtd></mtr></mtable><mo>,</mo></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>16</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0006.tif" /><br /> where TR2Snoise is the minimum target SNR between speech and ambient background noise, and mxG_AVBis a maximum allowable AVB gain, e.g. 20 dB. In order to change the AVB gain gradually, in one embodiment it is constrained to change in small step sizes, and the actual AVB gain is calculated as

0120<maths id="MATH-US-00007" num="00007"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><msub><mi>G</mi><mi>AVB</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mi>k</mi><mo>)</mo></mrow></mrow><mo>=</mo><mrow><mo>{</mo><mrow><mtable><mtr><mtd><mrow><mrow><msub><mi>G</mi><mi>AVB</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>-</mo><mn>1</mn></mrow><mo>)</mo></mrow></mrow><mo>+</mo><mi>Δ</mi></mrow></mtd><mtd><mrow><msub><mi>TG</mi><mi>AVB</mi></msub><mo>></mo><mrow><mrow><msub><mi>G</mi><mi>AVB</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>-</mo><mn>1</mn></mrow><mo>)</mo></mrow></mrow><mo>+</mo><mi>Δ</mi></mrow></mrow></mtd></mtr><mtr><mtd><mrow><mrow><msub><mi>G</mi><mi>AVB</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>-</mo><mn>1</mn></mrow><mo>)</mo></mrow></mrow><mo>-</mo><mi>Δ</mi></mrow></mtd><mtd><mrow><msub><mi>TG</mi><mi>AVB</mi></msub><mo><</mo><mrow><mrow><msub><mi>G</mi><mi>AVB</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>-</mo><mn>1</mn></mrow><mo>)</mo></mrow></mrow><mo>-</mo><mi>Δ</mi></mrow></mrow></mtd></mtr><mtr><mtd><mrow><msub><mi>G</mi><mi>AVB</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>-</mo><mn>1</mn></mrow><mo>)</mo></mrow></mrow></mtd><mtd><mi>otherwise</mi></mtd></mtr></mtable><mo>,</mo></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>17</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0007.tif" /><br /> where Δ is the step size, e.g. 1 dB.

0121With respect to the minimum target SNR, in practice a value of 15 dB may work in an embodiment in which the telephony terminal is a hanging style Bluetooth® headset. However, it is anticipated that the specific value will depend somewhat on the actual telephony terminal implementation. For example, an alternative Bluetooth® headset having an in-ear style speaker that provides a good acoustic seal will prevent some of the ambient background noise from reaching the auditory system of the user. In that case, a lower minimum SNR such as 6 dB may work. If the attenuation by the seal is accounted for in the calculations in the algorithm, e.g. the SNR is specified at the point of the ear drum, then the desired minimum SNR should be more device independent. However, in practice it may not be simple to account for such factors as the seal.

0122The receive-signal-to-ambient-background-noise ratio is a key parameter that is monitored by SIE system <b>200</b>. Note that the far-end speech signal and the near-end noise are two different signals in two different domains. Even for the same far-end speech signal level as “seen” by SIE system <b>200</b>, different loudness levels may be perceived by the user of the near-end telephony terminal depending on the gain applied to the speech signal before playback, the loudspeaker sensitivity, and a number of other factors. Similarly, even for the same near-end background noise level in the acoustic domain, SIE system <b>200</b> may see different noise levels depending on the microphone sensitivity, the gain applied to the microphone signal, or the like. Therefore, it is anticipated that for each type of telephony terminal, some calibration will be needed so that the predetermined SNR target as measured by the SIE system and method makes sense.

0123After the actual AVB gain has been calculated, AVB logic <b>224</b> then calculates the desired total gain as <br /><i>G</i>_desired=volume+<i>G</i>_AGC<i>+G</i>_AVB,  (18)<br /> where volume is the user volume of the telephony terminal (set by the user). Depending upon the implementation, there could be an additional term corresponding to a loss dictated by an echo suppression algorithm. This term is shown as “receive suppression” in <figref idref="DRAWINGS">FIGS. 2 and 8</figref> and may be received, for example, from a sub-band acoustic echo cancellation (SBAEC) component or other component within the telephony device.

0124Compression logic <b>226</b> then computes the final gain, wherein the final gain represents any compression that will be applied. The instant attack of the waveform envelope tracking as described above in reference to Eq. 13 taken together with the following gain calculations essentially guarantees that saturation and clipping will never occur.

0125To compute the final gain, compression logic <b>226</b> first calculates a compression point, C_p, relative to maximum digital amplitude in a manner that is adaptive and that takes into account the user volume and a calibration value for a “nominal” user (at a nominal listening level in quiet): <br /><i>C</i>_p=max└C_p,default_{_}_volume+(default_volume−volume),0,  (19)<br /> where C_p,default_{_}_volumeis the compression point at a user volume of default_volume. One can think of C_p,default_{_}_volumeas the maximum comfortable waveform level for a user that would use default_volume in quiet.

0126This adaptive approach to determining the compression point advantageously allows the compression point to move up and down with the user volume. For example, a compression point of 6 dB means that compression logic <b>226</b> will limit the waveform amplitude to 6 dB below maximum digital amplitude. For a user who prefers and uses a higher volume compared to another user, this means that compression point will be closer to maximum digital amplitude, and hence the signal will compressed at a higher level allowing higher waveform levels. For a user with a 3 dB louder volume setting, the compression will occur at a waveform amplitude that is 3 dB higher.

0127In further accordance with this approach, the waveform amplitude will be limited by compression logic <b>226</b> to a level that is below the maximum digital amplitude, and hence the full digital range may not be utilized for some users. In cases where this is undesirable, the compression point could be fixed to 0 dB. For example, this could apply to telephony terminals that are unable to provide sufficient volume for any user. However, where a telephony terminal is capable of providing more than enough loudness for a user (i.e., the loudness can be increased to a point that is uncomfortable for the user), the above approach of adaptively determining the compression point ensures that a level of discomfort will not be exceeded. Instead, loudness is achieved by amplifying the lower level segments while preventing the higher level segments from exceeding the compression point, which can be viewed as representing the maximum waveform amplitude of comfort.

0128Consequently, using this adaptive approach to determine the compression point, a higher maximum waveform is allowed for a user with a higher user volume setting, acknowledging that this particular user prefers louder levels. Conversely, a user with high sensitivity applying a lower user volume setting will be protected by a lower compression point (further below maximum digital amplitude). Instead of achieving intelligibility by uncomfortable levels via linear gain, the intelligibility is achieved by the additional features such as amplification of lower levels, spectral shaping, and dispersion.

0129In some sense, the adaptive nature of the compression point offers acoustic shock protection to users by limiting the maximum amplitude of waveforms that the auditory system is exposed to. The use of such a compression point also means that sometimes the maximum possible linear gain is not applied, and instead intelligibility is achieved by other means in order to honor the user's sensitivity to pure linear gain. Hence, in the interest of avoiding user discomfort, processing that introduces distortion in a traditional sense may be activated before distortion-less processing (linear gain) has been exhausted. However, from the perspective of the auditory system of the user the discomfort can be considered a distortion, and hence the above-described application of processing that increases distortion in a traditional sense should not be considered a violation of the prescribed philosophy of applying increasingly more aggressive processing as noise levels increase.

0130Furthermore, not only does the adaptive compression point accommodate users with different sensitivity, it also accommodates a varying acoustic seal for a single user. This frequently occurs when the user is using a cellular telephone, Bluetooth® headset, or like device that is often coupled and uncoupled from the ear, acoustically. If the seal is 3 dB worse during one use, the user would naturally increase volume by 3 dB to achieve the same loudness. Consequently, the compression point will move up by 3 dB, and everything will behave as before. As can be seen from Eq. 19 the compression point is not allowed to go beyond 0 dB, i.e. the maximum digital amplitude. This, along with the instant attack of the waveform tracking, prevents any kind of saturation.

0131It should be noted that in some cases it may be beneficial to allow some digital saturation since this will also provide some additional loudness. In terms of determining the permissible amount of saturation, a suitable trade-off must be made between loudness and distortion from saturation. As described in commonly-owned, co-pending U.S. patent application Ser. No. 12/109,017 (entitled “Audio Signal Shaping for Playback by Audio Devices” and filed Apr. 24, 2008), the entirety of which is incorporated by reference herein, soft-clipping may be used to minimize objectionable distortion. In such an implementation, the threshold in Eq. 19 will not be 0, but rather a negative number with an absolute value corresponding to the acceptable level of clipping.

0132After the compression point C_phas been determined, compression logic <b>226</b> calculates the overall gain headroom, G_headroom, between the waveform and the compression point as

0133<maths id="MATH-US-00008" num="00008"><math overflow="scroll"><mtable><mtr><mtd><mrow><msub><mi>G</mi><mi>headroom</mi></msub><mo>=</mo><mrow><mrow><mn>20</mn><mo>·</mo><mrow><msub><mi>log</mi><mn>10</mn></msub><mo>(</mo><mfrac><mi>MAXAMPL</mi><mrow><mi>mx</mi><mo>⁡</mo><mrow><mo>(</mo><mi>k</mi><mo>)</mo></mrow></mrow></mfrac><mo>)</mo></mrow></mrow><mo>-</mo><msub><mi>G</mi><mi>margin</mi></msub><mo>-</mo><msub><mi>C</mi><mi>p</mi></msub></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>20</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0008.tif" /><br /> where MAXAMPL is the maximum digital amplitude of the output in the system, e.g. 32768 for a 16-bit output. The gain headroom is calculated as the gain required to bring the waveform envelope tracking information, denoted mx(k), to the compression point, or just below if a margin, G_margin, is desired due to finite precision of fixed point arithmetic, e.g. G_margin=1 dB. In the special case where the compression point is 0 dB, and hence corresponds to the point of saturation, the gain headroom corresponds to the headroom between the waveform envelope and saturation, less the margin, G_margin.

0134Compression logic <b>226</b> then calculates the final gain, G_final, to be applied to the current frame as the minimum of the desired linear gain and the gain headroom (observing the compression point). The time-varying final gain creates the compression effect due to lower level frames having greater gain headroom than higher level frames. <br /><i>G</i>_final=min[<i>G</i>_desired<i>,G</i>_headroom],  (21)<br /> Compression logic <b>226</b> then converts the final gain G_finalfrom the log domain to the linear domain <br /><i>g=</i>10^G^{<sub2>final</sub2>}^/20  (22)<br /> and gain application module <b>802</b> applies the converted gain g to the output signal from spectral shaping block <b>202</b>/dispersion filter <b>204</b> to produce the output signal (denoted “receive-out” in <figref idref="DRAWINGS">FIGS. 1 and 2</figref>) for playback via a loudspeaker of the telephony terminal: <br /><i>r</i>_out(<i>n</i>)=<i>g·y</i>(<i>n</i>).  (23)

01356. Compression Tracker <b>210</b>

0136Compression tracker <b>210</b> is configured to monitor the shortage in headroom, or instantaneous volume loss <br /><i>V</i>_instloss<i>=G</i>_desired<i>−G</i>_final,  (24)<br /> and to calculate an average version according to the following equations. First a peak tracker is updated according to

0137<maths id="MATH-US-00009" num="00009"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><msub><mi>V</mi><mi>peakloss</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mi>k</mi><mo>)</mo></mrow></mrow><mo>=</mo><mrow><mo>{</mo><mrow><mtable><mtr><mtd><msub><mi>V</mi><mi>linstoss</mi></msub></mtd><mtd><mrow><msub><mi>V</mi><mi>linstoss</mi></msub><mo>></mo><mrow><msub><mi>V</mi><mi>peakloss</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mi>k</mi><mo>)</mo></mrow></mrow></mrow></mtd></mtr><mtr><mtd><mrow><mrow><mn>4095</mn><mo>/</mo><mn>4096</mn></mrow><mo>·</mo><mrow><msub><mi>V</mi><mi>peakloss</mi></msub><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>k</mi><mo>-</mo><mn>1</mn></mrow><mo>)</mo></mrow></mrow></mrow></mtd><mtd><mi>otherwise</mi></mtd></mtr></mtable><mo>.</mo></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mn>25</mn><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US9373339B2_D0009.tif" /><br /> Then, compression tracker <b>210</b> applies second order smoothing to calculate the smoothed volume loss <br /><i>V</i>_loss(<i>k</i>)=2<i>β·V</i>_loss(<i>k</i>−)−β²<i>·V</i>_loss(<i>k−</i>2)+1<i>/β·V</i>_peakloss(<i>k</i>).  (26)<br /> where β is a smoothing factor, e.g. β=1023/1024. Compression tracker <b>210</b> feeds back the smoothed volume loss V_loss(k) back to spectral shaping block <b>202</b> and optionally dispersion filter <b>204</b> to control the operation thereof.

C. Alternate System Implementation

0138<figref idref="DRAWINGS">FIG. 9</figref> is a block diagram of an example SIE system <b>900</b> in accordance with an alternate embodiment of the present invention. Like SIE system <b>200</b> described above in reference to <figref idref="DRAWINGS">FIG. 2</figref>, SIE system <b>900</b> is configured to improve the intelligibility of a speech signal received over a communication network from a far-end telephony terminal (the “far-end speech signal”) for playback by a near-end telephony terminal of which SIE system <b>900</b> is a part. In <figref idref="DRAWINGS">FIG. 9</figref>, the far-end speech signal is denoted “Receive in.” The output of SIE system <b>900</b> is the modified far-end speech signal, denoted “Receive out.”

0139As shown in <figref idref="DRAWINGS">FIG. 9</figref>, SIE system <b>900</b> includes a first level estimator <b>902</b>, a dynamic filtering block <b>904</b>, a second level estimator <b>906</b>, AGC logic <b>908</b>, AVB logic <b>910</b>, suppression logic <b>912</b>, compression logic <b>914</b>, acoustic shock protection (ASP) logic <b>916</b>, a volume application block <b>918</b> and a soft clipper <b>920</b>. Each of these elements will now be described.

0140First level estimator <b>902</b> is configured to determine an estimated level of the far-end speech signal and to provide this information to AGC logic <b>908</b> and AVB logic <b>910</b> for use in performing gain calculations. By performing level estimation directly on the original far-end speech signal (as opposed to the far-end speech signal after processing by dynamic filtering block <b>904</b>), first level estimator <b>902</b> is able to provide AGC logic <b>908</b> and AVB logic <b>910</b> with a more accurate estimate of the level of the far-end speech signal as received from the communication network. However, in accordance with this implementation, first level estimator <b>902</b> cannot take into account any loss in level due to the processing of the far-end speech signal by dynamic filtering block <b>904</b>. In contrast, if level estimation for the purposes of performing AGC and AVB operations were performed after dynamic filtering, this could lead to the removal of the higher intensity components (i.e., lower-frequency components) which have less impact on intelligibility or loudness. In either case, one could include logic to compensate for the loss of loudness due to the operations of dynamic filtering block <b>904</b> to provide a more accurate estimate of the final loudness that a user would experience.

0141Dynamic filtering block <b>904</b> is configured to filter the far-end speech signal in an adaptive manner in order to increase intelligibility of the signal and to obtain more digital headroom for boosting of the signal by AVB logic <b>910</b> while avoiding the introduction of an impermissible level of digital saturation. The operations performed by dynamic filtering block <b>904</b> may include any of the functions attributed to spectral shaping block <b>202</b> and/or dispersion filter <b>204</b> as described above in reference to system <b>200</b> of <figref idref="DRAWINGS">FIG. 2</figref>. In an embodiment in which dynamic filtering block <b>904</b> performs spectral shaping and/or dispersion filtering, the degree of spectral shaping or dispersion filtering applied may be controlled by a measure of the amount of compression applied by compression logic <b>914</b> and/or ASP logic <b>916</b> or by a measure of the amount of digital headroom remaining before such compression will be applied.

0142In alternate implementations, the degree of spectral shaping or dispersion filtering applied may be a function of a long-term or average level of the far-end speech signal or as a function of the level of lower-frequency components of the far-end speech signal. The level of such lower-frequency components may be determined, for example, by passing the far-end speech signal through a low-pass filter that has a roughly inverse shape to a high-pass filter used by dynamic filtering block <b>904</b>.

0143Second level estimator <b>906</b> is configured to determine an estimated level of the far-end speech signal after it has been processed by dynamic filtering block <b>906</b>. This estimate is then provided to suppression logic <b>912</b>, compression logic <b>914</b> and ASP logic <b>916</b> for use in calculations performed by those blocks.

0144AGC logic <b>908</b> is configured to compensate for variations in the level of the far-end speech signal, as estimated by first level estimator <b>902</b>. For example, such variations may be due to variation of network connections, acoustic coupling, or the like. AGC logic <b>908</b> calculates a gain scaling factor that, when applied to the far-end speech signal, brings the far-end speech signal to a nominal signal level. AGC logic <b>908</b> may operate in a like manner to that described above in reference to AGC logic <b>222</b> of system <b>200</b> or in a manner to be described below in Section D.

0145AVB logic <b>910</b> is configured to calculate an additional gain to be applied to the far-end speech signal so as to maintain a minimum SNR between the level of the far-end speech signal (after application of the gain calculated by AGC logic <b>908</b>) and the level of the near-end background noise. AVB logic <b>910</b> may operate in a like manner to that described above in reference to AVB logic <b>224</b> of system <b>220</b>.

0146Suppression logic <b>912</b> is configured to apply an echo suppression algorithm to the far-end speech signal in order to attenuate the effects of acoustic echo on that signal. The output of suppression logic <b>912</b> is a loss to be applied to the far-end speech signal.

0147Compression logic <b>914</b> is configured to determine a time varying gain to be applied to the far-end speech signal to ensure that, after application of the gain calculated by AGC logic <b>908</b>, the gain calculated by AVB logic <b>910</b>, the gain calculated by suppression logic <b>912</b>, and a gain associated with a user volume setting, the audio output waveform does not exceed (or exceeds by only a permissible amount) a digital saturation or clipping point of the telephony device.

0148ASP logic <b>916</b> is configured to adaptively determine a compression point (i.e., an offset from a maximum digital amplitude at which saturation occurs) below which the maximum amplitude of the far-end speech signal must be maintained in order to protect users of the telephony device from acoustic shock or discomfort. ASP logic <b>916</b> may thus be thought of as calculating an additional loss that must be applied to the far-end speech signal in addition to that determined by compression logic <b>914</b>.

0149Volume application block <b>918</b> is configured to receive the far-end speech signal after processing by dynamic filtering block <b>904</b> and to apply the gains calculated by AGC logic <b>908</b>, AVB logic <b>910</b>, suppression logic <b>912</b>, compression logic <b>914</b> and ASP logic <b>916</b>, as well as a gain associated with a user volume, thereto.

0150Soft clipper <b>920</b> is configured to receive the audio signal output by volume application block <b>918</b> and apply soft clipping thereto. Soft clipper <b>920</b> operates by manipulating the dynamic range of the audio signal output by volume application block <b>918</b> such that the level of the signal does not exceed a soft clipping limit. The soft clipping limit may be less than a limit imposed by the compression logic <b>914</b>/ASP logic <b>916</b>. In accordance with such an embodiment, at higher volumes, the dynamic range of the audio signal output by volume application block <b>918</b> will exceed the soft clipping limit of soft clipper <b>920</b>. This overdriving of soft clipper <b>920</b> will lead to some level of clipping distortion. However, through careful selection of the limit imposed by compression logic <b>914</b>/ASP logic <b>916</b> and the soft clipping limit, the amount of clipping distortion can advantageously be held to an acceptable level while maintaining loudness. An example of the use of soft clipping subsequent to amplitude compression is described in previously-referenced U.S. patent application Ser. No. 12/109,017, the entirety of which is incorporated by reference herein.

D. Alternate AGC Logic Implementation

0151<figref idref="DRAWINGS">FIG. 10</figref> is a block diagram of AGC logic <b>1000</b> that may be used to implement AGC logic <b>222</b> of SIE system <b>200</b> (described above in reference to <figref idref="DRAWINGS">FIG. 2</figref>) or AGC logic <b>908</b> of SIE system <b>900</b> (described above in reference to <figref idref="DRAWINGS">FIG. 9</figref>) in accordance with alternate embodiments of the present invention.

0152As shown in <figref idref="DRAWINGS">FIG. 10</figref>, AGC logic <b>1000</b> includes a long-term level estimator <b>1002</b>. Long-term level estimator <b>1002</b> is configured to periodically receive a short-term estimate of the level of the far-end speech signal and to update a long-term estimate of the level of the far-end speech signal based on the short-term level estimate. With reference to system <b>900</b> of <figref idref="DRAWINGS">FIG. 9</figref>, the short-term level estimate may be received from level estimator <b>902</b>.

0153A combiner <b>1004</b> is configured to receive the long-term level estimate generated by long-term level estimator <b>1002</b> and to add a current AGC gain thereto. The output of this operation is provided to decision logic <b>1006</b>.

0154Decision logic <b>1006</b> determines whether or not the output of combiner <b>1004</b> exceeds a target level. If the output exceeds the target level, then a logic block <b>1008</b> operates to adjust the current AGC gain downward so that the target level can be maintained. Conversely, if the output does not exceed the target level, then a logic block <b>1010</b> operates to adjust the current AGC gain upward so that the target level can be maintained. Note that in certain embodiments, the target level may be a configurable parameter.

0155In an embodiment, long-term level estimator <b>1002</b> is also configured to receive a “receive active” signal from a sub-band acoustic echo canceller (SBAEC) that indicates whether or not the far-end speech signal constitutes active speech as well as a “send active” signal from the SBAEC that indicates whether or not a near-end speech signal to be transmitted to a far-end telephony device constitutes active speech. In a circumstance in which both the “receive active” and “send active” signals are asserted, long-term level estimator <b>1002</b> will not reduce the long-term level estimate it produces regardless of the short-term level estimates received (i.e., the long-term level estimate will not be allowed to adapt downward). The net result of this will be that the magnitude of the AGC gain will not be adapted upward even if the short-term level estimates are decreasing. This feature is intended to ensure that AGC logic <b>1000</b> does not operate to undo a loss that may be applied to the far-end speech signal by an echo suppressor when both the “receive active” and “send active” signals are asserted.

0156However, when both the “receive active” and “send active” signals are asserted, long-term level estimator <b>1002</b> will remain capable of increasing the long-term level estimate that it produces based on the short-term level estimates received (i.e., the long-term level estimate is allowed to adapt upward). This ensures that the AGC gain can still be adapted downward to maintain the target signal level if the far-end speech signal is too loud.

0157In an embodiment, AVB logic that operates in conjunction with AGC logic <b>1000</b> (e.g., AVB logic <b>224</b> of system <b>200</b> or AVB logic <b>910</b> of system <b>900</b>) is configured to determine the amount of AVB gain to be applied to the far-end speech signal based also on a long-term level estimate that is not allowed to adapt downward when both the near-end speech signal and the far-end speech signal are determined to constitute active speech. This ensures that the AVB logic also does not operate to undo echo suppression that may have been applied to the far-end speech signal. However, the long-term level estimate used by the AVB logic is allowed to adapt upward when both the near-end speech signal and the far-end speech signal are determined to constitute active speech.

0158In a further embodiment, long-term level estimator <b>1002</b> is capable of determining whether the far-end speech signal constitutes tones or stationary (i.e., non-speech) signals based on an analysis of the short-term level estimate. In further accordance with such an embodiment, if it is determined that the far-end speech signal constitutes tones or stationary signals, long-term level estimator <b>1002</b> will prevent the long-term level estimate from adapting downward but allow the long-term level estimate to adapt upwards in a like-manner to that described above when both the when both the “receive active” and “send active” signals are asserted.

0159Note that in one implementation, the compression point used for applying amplitude compression (as previously described) can be made adaptive such that a different compression point is used when the “send active” signal is asserted (which may be indicative of doubletalk) or when the far-end speech signal is determined to constitute tones or stationary signals.

E. Example Integration with Sub-band Acoustic Echo Canceller

0160<figref idref="DRAWINGS">FIG. 11</figref> is a block diagram that shows a telephony terminal <b>1100</b> in which an SIE system in accordance with an embodiment of the present invention is integrated with a sub-band acoustic canceller. As shown in <figref idref="DRAWINGS">FIG. 11</figref>, telephony terminal <b>1100</b> includes a receive processing block <b>1102</b> that is configured to improve the intelligibility of a speech signal received over a communication network from a far-end telephony terminal (the “far-end speech signal”) for playback by telephony terminal <b>1100</b>. In <figref idref="DRAWINGS">FIG. 11</figref>, the far-end speech signal is denoted “Receive in.” The output of receive processing block <b>1102</b> is the modified far-end speech signal, denoted “Receive out.” Receive processing block <b>1102</b> includes an SIE system in accordance with an embodiment of the present invention, such as SIE system <b>200</b> described above in reference to <figref idref="DRAWINGS">FIG. 2</figref> or SIE system <b>900</b> described above in reference to <figref idref="DRAWINGS">FIG. 9</figref>.

0161As further shown in <figref idref="DRAWINGS">FIG. 11</figref>, telephony terminal <b>1100</b> includes a sub-band acoustic canceller <b>1104</b> that operates to cancel acoustic echo present in a speech signal captured by telephony terminal <b>1100</b> for transmission to the far-end telephony terminal over the communication network (the “near-end speech signal”). In <figref idref="DRAWINGS">FIG. 11</figref>, the near-end speech signal is denoted “Send in.” The output of sub-band acoustic echo canceller <b>1104</b> is the modified near-end speech signal, denoted “Send out.”

0162Sub-band acoustic canceller <b>1104</b> includes a number of components including a first sub-band analysis block <b>1112</b>, a second sub-band analysis block <b>1114</b>, a sub-band cancellation block <b>1116</b>, a combiner <b>1118</b>, a receive estimation block <b>1120</b>, a send estimation block <b>1122</b>, a post processing block <b>1124</b> and a sub-band synthesis block <b>1126</b>. The operation of each of these components will now be described.

0163First sub-band analysis block <b>1112</b> is configured to receive a time-domain version of the near-end speech signal and to convert the signal into a plurality of frequency sub-band components. First sub-band analysis block <b>1112</b> may also down-sample the near-end speech signal as part of this process. Second sub-band analysis block <b>1114</b> is configured to receive a time-domain version of the modified far-end speech signal output by receive processing block <b>1102</b> and to convert the signal into a plurality of frequency sub-band components. First sub-band analysis block <b>1112</b> may also down-sample the near-end speech signal as part of this process.

0164Sub-band cancellation block <b>1116</b> receives the sub-band representation of the near-end speech signal and the modified far-end speech signal and operates to determine, on a sub-band by sub-band basis, components of the near-end speech signal that represent acoustic echo and thus should be cancelled from the signal. To perform this function, sub-band cancellation block <b>1116</b> analyzes the level of correlation between the near-end speech signal and the modified far-end speech signal. The sub-band echo components are provided to a combiner <b>1118</b> which operates to subtract the echo components from the near-end speech signal on a sub-band by sub-band basis.

0165Post processing block <b>1124</b> is configured to receive the signal output by combiner <b>1118</b> and to perform non-linear processing thereon to remove residual echo as well as to perform processing thereon to suppress noise present in the signal.

0166Sub-band synthesis block <b>1126</b> is configured to receive the output from post processing block <b>1124</b>, which is represented as a plurality of frequency sub-band components, and to convert the plurality of sub-band components into a time domain representation of a modified version of the near-end speech signal. Sub-band synthesis block <b>1126</b> may also up-sample the modified version of the near-end speech signal as part of this process. The modified version of the near-end speech signal produced by sub-band synthesis block <b>1126</b> is then output for encoding and subsequent transmission to the far-end telephony terminal over the communication network.

0167Receive estimation block <b>1120</b> is configured to receive the sub-band components of the modified far-end speech signal and to estimate levels associated with each of the sub-bands that are used by sub-band cancellation block <b>1116</b> for performing acoustic echo cancellation functions and by post processing block <b>1124</b> for performing non-linear processing and noise suppression. The estimated levels may include, for example, an estimated level of a speech signal component present within each sub-band, an estimated level of a noise component present within each sub-band, or the like.

0168Send estimation block <b>1122</b> is configured to receive the sub-band components of the near-end speech signal after echo cancellation and to estimate levels associated with each of the sub-bands that are used by sub-band cancellation block <b>1116</b> for performing acoustic echo cancellation functions and by post processing block <b>1124</b> for performing non-linear processing and noise suppression. The estimated levels may include, for example, an estimated level of a speech signal component present within each sub-band, an estimated level of a noise component present within each sub-band, or the like.

0169In accordance with an embodiment of the present invention, sub-band acoustic canceller <b>1104</b> provides certain information generated during the performance of echo cancellation and noise suppression operations to receive processing block <b>1102</b>. Receive processing block <b>1102</b> then uses such information to perform SIE operations. Such information will now be described.

0170In one embodiment, sub-band acoustic canceller <b>1104</b> provides a measure of voice activity in the far-end speech signal to one or more level estimator(s) in receive processing block <b>1102</b>. The measure of voice activity may be used to control the level estimation function. The measure of voice activity may be determined, for example, by counting the number of sub-bands in which the energy significantly exceeds a noise floor. Because sub-band acoustic canceller <b>1104</b> analyzes the far-end speech signal in sub-bands, it is capable of providing a more accurate measure of voice activity than an analysis of a time-domain signal would provide.

0171In a further embodiment, sub-band acoustic canceller <b>1104</b> also provides a measure of voice activity in the near-end speech signal to one or more level estimator(s) in receive processing block. This measure of voice activity may also be used to control the level estimation function. For example, as described in Section D, above, AGC logic within receive processing block <b>1102</b> may use a measure of the voice activity in the far-end speech signal and in the near-end speech signal to prevent upward adaption of a long-term level estimate when both the far-end speech signal and the near-end speech signal are deemed to constitute speech.

0172In another embodiment, sub-band acoustic canceller <b>1104</b> provides an estimate of the noise level present in the near-end speech signal to receive processing block <b>1102</b>. For example, AVB logic within receive processing block <b>1102</b> may receive an estimate of the noise level present in the near-end speech signal from sub-band acoustic canceller <b>1104</b> and use this estimate to determine a far-end speech signal to near-end noise ratio as previously described.

0173Since sub-band acoustic canceller <b>1104</b> estimates noise levels on a frequency sub-band basis, the estimate of the noise level present in the near-end speech signal may be determined by assigning greater weight to certain sub-bands as opposed to others in order to ensure that the estimated noise level represents noise that would be perceptible to a human (in other words to ensure that the estimated noise level is a measure of the loudness of the noise as opposed to the intensity).

0174Furthermore, since sub-band acoustic canceller <b>1104</b> estimates noise levels on a frequency sub-band basis, sub-band acoustic canceller <b>1104</b> can provide the sub-band noise level estimates to a spectral shaping block within receive processing block <b>1102</b>, such that spectral shaping may be performed as a function of the spectral shape of the noise. For example, different spectral shaping may be applied when the noise is white as opposed to flat.

0175It is noted that in <figref idref="DRAWINGS">FIG. 11</figref>, the speech signals denoted “Receive in,” “Receive out,” “Send in” and “Send out” are represented using two lines. This is intended to indicate that telephony terminal <b>1100</b> is capable of processing wideband speech signals (e.g., signals generated using 16 kHz sampling). In one embodiment of telephony terminal <b>1100</b>, the far-end and near-end speech signals are wideband speech signals and are split into a narrowband component (e.g., 0-3.4 kHz, sampled at 8 kHz) and a wideband component (e.g., 3.4-7 kHz, sampled at 16 kHz). This approach makes the signal processing aspects of the terminal simpler from a wideband/narrowband perspective and enables functionality that is applicable only to narrowband speech signals to be implemented by processing only the narrowband component. Examples of systems that perform such split-band processing are described in U.S. Pat. Nos. 6,848,012, 6,928,495, 7,165,130, 7,283,585, 7,333,475 and 7,409,056 and U.S. patent application Ser. No. 11/672,120, the entireties of which are incorporated by reference herein.

0176In one embodiment, the SIE processing described above is applied only to a narrowband component of a wideband speech signal. In an alternate embodiment, the previously-described SIE processing is made applicable to wideband speech by also modifying the wideband component of a wideband speech signal. For example, in one embodiment, the gain of filters used to modify the far-end speech signal by receive processing block <b>1102</b> at 3.4 kHz (or 4 kHz) are extended across the wideband component. In slowly evolving spectral shaping, a table of the gain for the wideband component may be utilized, wherein the gain is a function of the narrowband filter. In one implementation, for rapidly evolving spectral shaping, the gain of the filter at 4 kHz is unity, so that there is no need to modify the wideband component.

0177The foregoing concept may also be extended to other sets of signal components sampled at various sampling rates, such as 8 kHz/16 kHz/48 kHz or 8 kHz/48 kHz.

0178<figref idref="DRAWINGS">FIG. 12</figref> is a block diagram that shows an alternate telephony terminal <b>1200</b> in which an SIE system in accordance with an embodiment of the present invention is integrated with a sub-band acoustic canceller. Telephony terminal <b>1200</b> differs from telephony terminal <b>1100</b> in a variety of ways.

0179For example, telephony terminal <b>1200</b> is configured to receive a plurality of speech signals, denoted “Receive in” <b>1</b> through “Receive in” m, and to combine those signals to produce a single output speech signal denoted “Receive out.” Each of the signals “Receive in” 1-m may comprise, for example and without limitation, a different far-end speech signal in a multi-party conference call or a different audio channel in a multi-channel audio signal.

0180As shown in <figref idref="DRAWINGS">FIG. 12</figref>, each “Receive in” signal 1-m is processed by a corresponding receive processing block <b>1202</b>₁-<b>1202</b>_m. Each receive processing block <b>1202</b>₁-<b>1202</b>_mincludes an SIE system in accordance with an embodiment of the present invention, such as SIE system <b>200</b> described above in reference to <figref idref="DRAWINGS">FIG. 2</figref> or SIE system <b>900</b> described above in reference to <figref idref="DRAWINGS">FIG. 9</figref>, and operates to improve the intelligibility of a corresponding “Receive in” signal.

0181As further shown in <figref idref="DRAWINGS">FIG. 12</figref>, the output signals of receive processing blocks <b>1202</b>₁-<b>1202</b>_mare combined prior to being received by a compression and soft clipping block <b>1204</b>. By separately applying SIE to each “Receive in” signal prior to mixing, telephony terminal <b>1200</b> ensures that each “Receive in” signal is modified only to the extent necessary to achieve a desired intelligibility for that signal. In other words, by separately applying SIE to each “Receive in” signal, one “Receive in” signal need not be distorted to improve the intelligibility of another “Receive in” signal.

0182Compression and soft clipping logic <b>1204</b> is configured to apply amplitude compression and/or soft clipping to the signal produced by the combination of the outputs of receive processing blocks <b>1202</b>₁-<b>1202</b>_m. Such amplitude compression and/or soft clipping may be applied to ensure that the signal produced by the combination of the outputs of receive processing blocks <b>1202</b>₁-<b>1202</b>_mdoes not exceed a digital saturation point or only exceeds the digital saturation point by a permissible amount. Note that in an alternate implementation, compression and soft clipping may be separately applied to each signal output from each of receive processing blocks <b>1202</b>₁-<b>1202</b>_mand then further applied to the signal produced by the combination of those outputs.

0183As also shown in <figref idref="DRAWINGS">FIG. 12</figref>, telephony terminal <b>1200</b> includes a sub-band acoustic canceller <b>1204</b> that operates to cancel acoustic echo present in a near-end speech signal captured by telephony terminal <b>1200</b> for transmission to a far-end telephony terminal over a communication network. To capture the near-end speech signal, telephony terminal includes a plurality of microphones, each of which produces a different input speech signal. These input speech signals are denoted “Send in” <b>1</b> through “Send in” n. Each input speech signal “Send in” 1-n is converted from a time domain signal into a plurality of frequency sub-band components by a corresponding sub-band analysis block <b>1212</b>₁-<b>1212</b>_m. The output from sub-band analysis blocks <b>1212</b>₁-<b>1212</b>_mare provided to a beamformer <b>1228</b> which performs spatial filtering operations on the output to attenuate unwanted undesired audio content. The output of beamformer <b>1228</b> is then treated as the near-end speech signal.

0184The remaining components of sub-band acoustic echo canceller <b>1206</b> operate in essentially the same manner as like-named components described above in reference to telephony terminal <b>1100</b> of <figref idref="DRAWINGS">FIG. 11</figref>. However, to perform an estimation of the level of the noise in the near-end speech signal, send estimation block <b>1222</b> may be configured to account for the noise-reducing effect of beamformer <b>1228</b>. In other words, the noise level estimate provided by send estimation block <b>1222</b> may be an estimate of the noise level at one of the multiple microphones.

0185Sub-band acoustic canceller <b>1204</b> provides certain information generated during the performance of echo cancellation and noise suppression operations to receive processing blocks <b>1202</b>₁-<b>1202</b>_m. Each of receive processing blocks <b>1202</b>₁-<b>1202</b>_mthen uses such information to perform SIE operations. The information provided may include, for example and without limitation, a measure of voice activity in the far-end speech signal, a measure of voice activity in the near-end speech signal, or an estimate of the noise level present in the far-end speech signal.

0186In the implementation described above in reference to <figref idref="DRAWINGS">FIG. 12</figref>, a plurality of received speech signals “Receive in” 1-m are combined to produce a single “Receive out” speech signal. However, persons skilled in the relevant art(s) will readily appreciate that the present invention encompasses other implementations in which a one or more received speech are processed to produce a plurality of “Receive out” speech signals 1-n. For example, in an embodiment in which the invention is implemented in a stereo headset or a stereo Voice over IP Protocol (VoIP) telephone, one or more received speech signals may be processed to produce 2 channels of output audio. Depending upon the specific implementation, receive processing and/or compression/soft-clipping may be performed on each received speech signal as well as upon combinations of such received speech signals to produce the desired output signals.

F. Example Methods in Accordance with Embodiments of the Present Invention

0187Example methods for processing a speech signal for playback by an audio device in accordance with various embodiments of the present invention will now be described in reference to flowcharts depicted in <figref idref="DRAWINGS">FIGS. 13-21</figref>.

0188In particular, <figref idref="DRAWINGS">FIG. 13</figref> depicts a flowchart <b>1300</b> of a method for processing a portion of a speech signal to be played back by an audio device in accordance with one embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 13</figref>, the method of flowchart <b>1300</b> begins at step <b>1302</b> in which a level of the speech signal is estimated. At step <b>1304</b>, a level of background noise is estimated. At step <b>1306</b>, a signal-to-noise ratio (SNR) is calculated based on the estimated level of the speech signal and the estimated level of the background noise. At step <b>1308</b>, an amount of gain to be applied to the portion of the speech signal is calculated based on at least a difference between a predetermined SNR and the calculated SNR. At step <b>1310</b>, the amount of gain is applied to the portion of the speech signal.

0189In one embodiment, performing step <b>1306</b> comprises calculating an automatic gain control (AGC) gain required to bring the estimated level of the speech signal to a predefined nominal level and then calculating the SNR based on the estimated level of the speech signal after application of the AGC gain thereto and the estimated level of the background noise. For example, as described elsewhere herein, this step may comprise calculating: <br /><i>R</i>2<i>S</i>noise=default_volume+<i>G</i>_AGC<i>+L</i>_R<i>+C−L</i>_Snoise,<br /> wherein R2Snoise is the calculated SNR, default_volume is a constant representing a default volume, G_AGCis the AGC gain, L_Ris the estimated level of the speech signal, L_Snoiseis the estimated level of the background noise and C is a calibration term.

0190In one embodiment, performing step <b>1308</b> comprises performing a number of steps. These steps include calculating a target gain as the difference between the predetermined SNR and the calculated SNR. Then, an actual gain is compared to the target gain, wherein the actual gain represents an amount of gain that was applied to a previously-received portion of the speech signal. If the target gain exceeds the actual gain by at least a fixed amount, then the amount of gain to be applied to the portion of the speech signal is calculated by adding the fixed amount of gain to the actual gain. However, if the target gain is less than the actual gain by at least the fixed amount, then the amount of gain to be applied to the portion of the speech signal is calculated by subtracting the fixed amount of gain from the actual gain.

0191In another embodiment, performing step <b>1308</b> comprises summing at least a user volume of the audio device, an amount of gain determined based on the difference between the predetermined SNR and the calculated SNR, and an amount of gain required to bring the estimated level of the speech signal to a predefined nominal level.

0192In a further embodiment, performing step <b>1308</b> comprises first calculating a desired gain to be applied to the portion of the speech signal based on at least the difference between the predetermined SNR and the calculated SNR. Then, a determination is made as to whether the application of the desired gain to the portion of the speech signal would cause a reference amplitude associated with the portion of the speech signal to exceed a predetermined amplitude limit. If it is determined that the application of the desired gain to the portion of the speech signal would cause the reference amplitude to exceed the predetermined amplitude limit, then an amount of gain to be applied to the portion of the speech signal is calculated that is less than the desired gain. For example, as described elsewhere herein, calculating an amount of gain to be applied to the portion of the speech signal that is less than the desired gain may comprise calculating <br /><i>G</i>_final=min[<i>G</i>_desired<i>,G</i>_headroom],<br /> wherein G_finalis the amount of gain to be applied to the portion of the speech signal, G_desiredis the desired gain and G_headroomis an estimate of the difference between the reference amplitude associated with the portion of the speech signal and the predetermined amplitude limit.

0193In further accordance with this embodiment, a difference may be calculated between the desired gain and the amount of gain to be applied to the portion of the speech signal. Spectral shaping may then be applied to at least one subsequently-received portion of the speech signal, wherein the degree of spectral shaping applied is based at least in part on the difference. Alternatively or additionally, dispersion filtering may be performed on at least one subsequently-received portion of the speech signal, wherein the degree of dispersion applied by the dispersion filtering is based at least in part on the difference.

0194<figref idref="DRAWINGS">FIG. 14</figref> depicts a flowchart <b>1400</b> of a method for processing a speech signal to be played back by an audio device in accordance with an embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 14</figref>, the method of flowchart <b>1400</b> begins at step <b>1402</b>, in which a level of background noise is estimated. At step <b>1404</b>, a linear gain is applied to the speech signal if a function of at least the estimated level of background noise meets a first condition. At step <b>1406</b>, a linear gain and compression are applied to the speech signal if the function of at least the estimated level of the background noise meets a second condition. At step <b>1408</b>, a linear gain, compression and spectral shaping are applied to the speech signal if the function of at least the estimated level of background noise meets a third condition.

0195In one embodiment, each of the first, second and third conditions is indicative of a need for a corresponding first, second and third degree of speech intelligibility enhancement, wherein the second degree is greater than the first degree and the third degree is greater than the second degree. The function based on at least the estimated level of background noise may comprise, for example, a signal-to-noise ratio (SNR) that is calculated based on an estimated level of the speech signal and the estimated level of the background noise.

0196Although it is not shown in <figref idref="DRAWINGS">FIG. 14</figref>, the method of flowchart <b>1400</b> may also include applying a linear, gain, compression and dispersion filtering to the speech signal if at least the estimated level of background noise meets a fourth condition.

0197<figref idref="DRAWINGS">FIG. 15</figref> depicts a flowchart <b>1500</b> of another method for processing a portion of a speech signal to be played back by an audio device in accordance with an embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 15</figref>, the method of flowchart <b>1500</b> begins at step <b>1502</b>, in which a reference amplitude associated with the portion of the speech signal is calculated. In one embodiment, calculating the reference amplitude comprises determining a maximum absolute amplitude of the portion of the speech signal. In another embodiment, calculating the reference amplitude comprises determining a maximum absolute amplitude of a segment of the speech signal that includes the portion of the speech signal and one or more previously-processed portions of the speech signal. In a further embodiment, calculating the reference amplitude comprises setting the reference amplitude equal to the greater of a maximum absolute amplitude associated with the portion of the speech signal and a product of a reference amplitude associated with a previously-processed portion of the speech signal and a decay factor.

0198At step <b>1504</b>, a first gain to be applied to the portion of the speech signal is received.

0199At step <b>1506</b>, compression is applied to the portion of the speech signal if the application of the first gain to the portion of the speech signal would cause the reference amplitude associated with the portion of the speech signal to exceed a predetermined amplitude limit. In one embodiment, the predetermined amplitude limit comprises a maximum digital amplitude that can be used to represent the speech signal. In an alternate embodiment, the predetermined amplitude limit comprises an amplitude that is a predetermined number of decibels above or below a maximum digital amplitude that can be used to represent the speech signal.

0200The method of flowchart <b>1500</b> may further include adaptively calculating the predetermined amplitude limit. In one embodiment, adaptively calculating the predetermined amplitude limit comprises adaptively calculating the predetermined amplitude limit based at least on a user-selected volume.

0201Depending upon the implementation, the application of compression in step <b>1506</b> may include applying a second gain to the portion of the speech signal that is less than the first gain, wherein the second gain is calculated as an amount of gain required to bring the reference amplitude associated with the portion of the speech signal to the predetermined amplitude limit. As described previously herein, calculating the second gain may comprise calculating:

0202<maths id="MATH-US-00010" num="00010"><math overflow="scroll"><mrow><msub><mi>G</mi><mi>headroom</mi></msub><mo>=</mo><mrow><mrow><mn>20</mn><mo>·</mo><mrow><msub><mi>log</mi><mn>10</mn></msub><mo>(</mo><mfrac><mi>MAXAMPL</mi><mrow><mi>mx</mi><mo>⁡</mo><mrow><mo>(</mo><mi>k</mi><mo>)</mo></mrow></mrow></mfrac><mo>)</mo></mrow></mrow><mo>-</mo><msub><mi>G</mi><mi>margin</mi></msub><mo>-</mo><msub><mi>C</mi><mi>p</mi></msub></mrow></mrow></math></maths><img file="US9373339B2_D0010.tif" /><br /> wherein G_headroomis the second gain, MAXAMPL is a maximum digital amplitude that can be used to represent the speech signal, mx(k) is the reference amplitude associated with the portion of the speech signal, G_marginis a predefined margin and C_pis a predetermined number of decibels.

0203At step <b>1508</b>, a value representative of an amount of compression applied to the portion of the speech signal during step <b>1506</b> is calculated. In one embodiment, calculating this value comprises calculating an instantaneous volume loss by determining a difference between the first gain and the second gain described in the previous paragraph and then calculating an average version of the instantaneous volume loss.

0204At step <b>1510</b>, spectral shaping and/or dispersion filtering is applied to at least one subsequently-received portion of the speech signal wherein the degree of spectral shaping and/or dispersion filtering applied is controlled at least in part by the value calculated during step <b>1508</b>.

0205<figref idref="DRAWINGS">FIG. 16</figref> depicts a flowchart <b>1600</b> of another method for processing a portion of a speech signal to be played back by an audio device in accordance with an embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 16</figref>, the method of flowchart <b>1600</b> begins at step <b>1602</b>, at which a portion of the speech signal is received.

0206At step <b>1604</b>, a degree of spectral shaping to be applied to the portion of the speech signal to increase the intelligibility thereof is adaptively determined. Various methods may be used to adaptively determine the degree of spectral shaping to be applied. For example, a degree of compression that was or is estimated to be applied to the speech signal may be determined and the degree of spectral shaping to be applied may be determined as a function of at least the degree of compression.

0207As another example, a level of the speech signal may be calculated and the degree of spectral shaping to be applied may be determined as a function of at least the level of the speech signal.

0208As still another example, a level of one or more sub-band components of the speech signal may be calculated and the degree of spectral shaping to be applied may be determined as a function of at least the level(s) of the sub-band component(s).

0209As a further example, a level of background noise may be estimated and the degree of spectral shaping to be applied may be determined as a function of at least the level of the background noise. Estimating the level of the background noise may comprise estimating a level of one or more sub-band components of the background noise and determining the degree of spectral shaping to be applied as a function of at least the estimated level of the background noise may comprise determining the degree of spectral shaping as a function of at least the level(s) of the sub-band component(s).

0210As a still further example, a spectral shape of the background noise may be determined and the degree of spectral shaping to be applied may be determined as a function of at least the spectral shape of the background noise.

0211At step <b>1606</b>, the determined degree of spectral shaping is applied to the portion of the speech signal. Applying the determined degree of spectral shaping to the portion of the speech signal may comprise amplifying at least one selected formant associated with the portion of the speech signal relative to at least one other formant associated with the portion of the speech signal. For example, applying the determined degree of spectral shaping to the portion of the speech signal may comprise amplifying a second and third formant associated with the portion of the speech signal relative to a first formant associated with the portion of the speech signal.

0212In one embodiment, applying the determined degree of spectral shaping to the portion of the speech signal comprises performing time-domain filtering on the portion of the speech signal using an adaptive high-pass filter.

0213Performing time-domain filtering on the portion of the speech signal using an adaptive high-pass filter may comprise performing time-domain filtering on the portion of the speech signal using a first adaptive spectral shaping filter and a second adaptive spectral shaping filter, wherein the second adaptive spectral shaping filter is configured to adapt more rapidly than the first adaptive spectral shaping filter. For example, the first adaptive spectral shaping filter may have the form <br /><i>x</i>(<i>n</i>)=<i>r</i>_in(<i>n</i>)−<i>b·r</i>_in(<i>n−</i>1)<br /> wherein x(n) is the output of the first adaptive spectral shaping filter, r_in(n) is the input to the first adaptive spectral shaping filter, and b is a filter coefficient that increases as a degree of compression that was or is estimated to be applied to the speech signal increases. In further accordance with this example, the second adaptive spectral shaping filter may have the form: <br /><i>y</i>(<i>n</i>)=<i>x</i>(<i>n</i>)−<i>c·x</i>(<i>n−</i>2)−<i>c·y</i>(<i>n</i>−1)<br /> wherein y(n) is the output of the second adaptive spectral shaping filter, x(n) is the input to the second adaptive spectral shaping filter and c is a control parameter. The control parameter c may be calculated based upon a degree of compression that was or is estimated to be applied to the speech signal. The control parameter c may also be calculated based upon a measure of a slope of a spectral envelope of the speech signal.

0214Alternatively, performing time-domain filtering on the portion of the speech signal using an adaptive high-pass filter may comprise using only the first adaptive spectral shaping filter described above or using only the second adaptive spectral shaping filter described above.

0215<figref idref="DRAWINGS">FIG. 17</figref> depicts a flowchart <b>1700</b> of another method for processing a portion of a speech signal to be played back by an audio device in accordance with an embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 17</figref>, the method of flowchart <b>1700</b> begins at step <b>1702</b> in which dispersion filtering is performed on the portion of the speech signal to reduce a magnitude of waveform peaks in the portion of the speech signal. At step <b>1704</b>, an amount of gain to be applied to the portion of the speech signal is increased responsive to the reduction of the magnitude of the waveform peaks in the portion of the speech signal.

0216In one embodiment, performing dispersion filtering on the portion of the speech signal as described in reference to step <b>1702</b> comprises reducing a peak-to-average ratio associated with the portion of the speech signal. Reducing a peak-to-average ratio associated with the portion of the speech signal may comprise, for example, reducing a peak-to-RMS ratio associated with the portion of the speech signal.

0217Performing dispersion filtering on the portion of the speech signal as described in reference to step <b>1702</b> may also comprise passing the portion of the speech signal through a fixed all-pass filter. The fixed all-pass filter may comprise, for example, a fixed sixth-order all-pass filter.

0218Alternatively, performing dispersion filtering on the portion of the speech signal as described in reference to step <b>1702</b> may comprise passing the portion of the speech signal through an adaptive all-pass filter. In accordance with such an embodiment, poles and zeros of the adaptive all-pass filter may be adapted based on local characteristics of the speech signal. For example, radii of the poles of the adaptive all-pass filter may be decreased during silence regions of the speech signal and increased during vowel regions of the speech signal. As another example, pole frequencies of the adaptive all-pass filter may be set to pole frequencies of a fixed all-pass filter during an initial portion of a voiced region of the speech signal and then the pole frequencies may be adapted during subsequent portions of the speech signal by tracking changes in the speech signal. Tracking changes in the speech signal may include estimating formant frequencies of the speech signal and guiding the adaptation of the pole frequencies of the all-pass filter based on the estimated formant frequencies. Tracking changes in the speech signal may also comprises performing a closed-loop pole frequency search to determine optimal pole frequencies and then guiding the adaptation of the pole frequencies of the all-pass filter based on the optimal pole frequencies. Performing the closed-loop pole frequency search to determine the optimal pole frequencies may comprise limiting the closed-loop pole frequency search to predefined search ranges around optimal pole frequencies associated with a previously-processed portion of the speech signal.

0219In another embodiment in which performing dispersion filtering on the portion of the speech signal comprises passing the portion of the speech signal through an adaptive all-pass filter, the adaptive all-pass filter may be adapted based on a value representative of an amount of compression applied to one or more previously-processed portions of the speech signal. Adapting the filter in this manner may include calculating a scaling factor based on the value representative of the amount of compression, wherein the scaling factor increases as the value increases, and then applying the scaling factor to radii of poles of the adaptive all-pass filter.

0220In a further embodiment, performing dispersion filtering on the portion of the speech signal as described in reference to step <b>1702</b> comprises passing the portion of the speech signal through an all-pass filter comprises selecting one of a collection of N all-pass filter candidates.

0221<figref idref="DRAWINGS">FIG. 18</figref> depicts a flowchart <b>1800</b> of another method for processing a portion of a speech signal to be played back by an audio device in accordance with an embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 18</figref>, the method of flowchart <b>1800</b> starts at step <b>1802</b> in which a degree of compression that was applied to at least one previously-received portion of the speech signal is determined. At step <b>1804</b>, dispersion filtering is performed on the portion of the speech signal, wherein the degree of dispersion applied by the dispersion filtering is based at least in part on the degree of compression that was applied to the at least one previously-received portion of the speech signal. Performing dispersion filtering in step <b>1804</b> may comprise, for example, passing the portion of the speech signal through an adaptive all-pass filter. The adaptive all-pass filter may be adapted based on a value representative of the degree of compression that was applied to the at least one previously-received portion of the speech signal. Adapting the filter in this manner may include calculating a scaling factor based on the value representative of the amount of compression, wherein the scaling factor increases as the value increases, and then applying the scaling factor to radii of poles of the adaptive all-pass filter.

0222<figref idref="DRAWINGS">FIG. 19</figref> depicts a flowchart <b>1900</b> of a method for operating an integrated speech intelligibility enhancement system and acoustic echo canceller in accordance with an embodiment of the present invention.

0223As shown in <figref idref="DRAWINGS">FIG. 19</figref>, the method of flowchart <b>1900</b> begins at step <b>1902</b> in which characteristics associated with a near-end speech signal to be transmitted by an audio device and/or a far-end speech signal received for playback by the audio device are calculated. Calculating the characteristics may include, for example, calculating an estimated level of background noise associated with the near-end speech signal. Calculating the estimated level of background noise associated with the near-end speech signal may include calculating an estimated level of background noise corresponding to each of a plurality of sub-band components of the near-end speech signal. Alternatively, calculating the estimated level of background noise associated with the near-end speech signal may comprise calculating a measure of loudness by applying a weight to one or more estimated levels of background noise corresponding to one or more sub-band components of the near-end speech signal.

0224At step <b>1904</b>, the far-end speech signal is modified based on at least the calculated characteristics to increase the intelligibility thereof. In an embodiment in which the calculated characteristics comprise one or more estimated levels of background noise corresponding to one or more sub-band components of the near-end speech signal, this step may comprise performing spectral shaping on the far-end speech signal based on one or more of the estimated levels of background noise corresponding to one or more of the sub-band components.

0225At step <b>1906</b>, acoustic echo present in the near-end speech signal is suppressed based on at least the calculated characteristics.

0226In one embodiment of the method of flowchart <b>1900</b>, calculating characteristics in step <b>1902</b> comprises determining whether voice activity is present in the far-end speech signal and modifying the far-end speech signal in step <b>1904</b> comprises controlling the operation of a level estimator based on the determination, wherein the level estimator calculates an estimated signal level associated with the far-end speech signal, and applying a gain to the far-end speech signal wherein the amount of gain applied is based on the estimated signal level. Determining whether voice activity is present in the far-end speech signal may comprise analyzing one or more sub-band components of the far-end speech signal.

0227In another embodiment of the method of flowchart <b>1900</b>, calculating characteristics in step <b>1902</b> comprises determining whether voice activity is present in the near-end speech signal and modifying the far-end speech signal in step <b>1904</b> comprises controlling the operation of a level estimator based on the determination, wherein the level estimator calculates an estimated signal level associated with the far-end speech signal, and applying a gain to the far-end speech signal wherein the amount of gain applied is based on the estimated signal level. Determining whether voice activity is present in the near-end speech signal may comprise analyzing one or more sub-band components of the near-end speech signal.

0228In a further embodiment of the method of flowchart <b>1900</b>, calculating characteristics in step <b>1902</b> comprises calculating the estimated level of background noise at one or more microphones in a plurality of microphones associated with the audio device. Calculating the estimated level of background noise at one or more microphones in the plurality of microphones associated with the audio device may comprise modifying an estimated level of background noise associated with the near-end speech signal to account for a noise changing effect produced by a beamformer coupled to the plurality of microphones.

0229<figref idref="DRAWINGS">FIG. 20</figref> depicts a flowchart <b>2000</b> of a method for processing first and second speech signals to produce an output speech signal for playback in accordance with an embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 20</figref>, the method of flowchart <b>2000</b> begins at step <b>2002</b> in which a portion of the first speech signal is received.

0230At step <b>2004</b>, the portion of the first speech signal is modified to increase the intelligibility thereof, wherein the degree of modification applied to the portion of the first speech signal is based at least on an estimated level of background noise.

0231At step <b>2006</b>, a portion of the second speech signal is received.

0232At step <b>2008</b>, the portion of the second speech signal is modified to increase the intelligibility thereof, wherein the degree of modification applied to the portion of the second speech signal is based at least on an estimated level of background noise.

0233At step <b>2010</b>, the modified portion of the first speech signal and the modified portion of the second speech signal to produce an output speech signal for playback.

0234The foregoing method of flowchart <b>2000</b> may further include applying amplitude compression to the output speech signal and/or applying soft clipping to the output speech signal. In the foregoing method of flowchart <b>2000</b>, step <b>2004</b> may include applying compression to the portion of the first speech signal to produce the modified portion of the first speech signal and/or applying soft clipping to the portion of the first speech signal to produce the modified portion of the first speech signal. Likewise, step <b>2008</b> may include applying compression to the portion of the second speech signal to produce the modified portion of the second speech signal and/or applying soft clipping to the portion of the second speech signal to produce the modified portion of the second speech signal.

0235<figref idref="DRAWINGS">FIG. 21</figref> depicts a flowchart <b>2100</b> of a method for updating an amount of gain to be applied to a first speech signal received for playback by an audio device in accordance with an embodiment of the present invention. As shown in <figref idref="DRAWINGS">FIG. 21</figref>, the method of flowchart <b>2100</b> begins at step <b>2102</b> in which it is determined whether a second speech signal to be transmitted from the audio device constitutes active speech. The results of the determination are analyzed during decision step <b>2104</b>.

0236If it is determined during decision step <b>2104</b> that the second speech signal does not constitute active speech, then the amount of gain is reduced in response to an increase in an estimated level of the first speech signal and the amount of gain is increased in response to a decrease in the estimated level of the first speech signal as shown at step <b>2106</b>. However, if it is determined during decision step <b>2104</b> that the second speech signal does constitute active speech, then the amount of gain is reduced in response to an increase in the estimated level of the first speech signal and the amount of gain is not increased in response to a decrease in the estimated level of the first speech signal as shown at step <b>2108</b>.

0237The method of flowchart <b>2100</b> may further include updating the estimated level of the first speech signal. Updating the estimated level of the first speech signal may include calculating a short-term estimate of the level of the first speech signal based on a received portion of the first speech signal and then updating a long-term estimate of the level of the first speech signal based on the short-term estimate. In accordance with such an embodiment, performing step <b>2108</b> of flowchart <b>2100</b> may comprise not decreasing the long-term estimate of the level of the first speech signal responsive to a decrease in the short-term estimate of the level of the first speech signal.

0238The method of flowchart <b>2100</b> may further include determining whether the first speech signal constitutes a tone and performing step <b>2108</b> responsive also to determining that the first speech signal constitutes a tone. The method of flowchart <b>2100</b> may still further include determining whether the first speech signal constitutes a stationary signal and performing step <b>2108</b> responsive also to determining that the first speech signal constitutes a stationary signal.

G. Example Waveforms Generated by Speech Intelligibility Enhancement System and Method in Accordance with Embodiments of the Present Invention

0239<figref idref="DRAWINGS">FIG. 22</figref> depicts a waveform plot <b>2200</b> of an exemplary far-end speech signal that may be processed by SIE system <b>200</b> as described above in reference to <figref idref="DRAWINGS">FIG. 2</figref>. For example, the far-end speech signal shown in plot <b>2200</b> may be the “Receive-in” signal that is received by spectral shaping block <b>202</b> in SIE system <b>200</b>. In further accordance with this example, <figref idref="DRAWINGS">FIGS. 23, 24 and 25</figref> depict waveform plots of corresponding output speech signals that may be produced by SIE system <b>200</b> responsive to processing the far-end speech signal shown in plot <b>2200</b> at different levels of ambient background noise.

0240In particular, <figref idref="DRAWINGS">FIG. 23</figref> depicts a waveform plot <b>2300</b> of a corresponding output speech signal produced by SIE system <b>200</b> when the level of ambient background noise is sufficient to trigger the application of AVB (i.e., when the level of ambient background noise is such that the far-end speech signal to near-end background noise ratio is less than the target minimum SNR even after the application of AGC) but is not sufficient to trigger amplitude compression. As shown in waveform plot <b>2300</b>, a pure linear gain has been applied to the far-end speech signal, thus resulting in a waveform having increased amplitude and loudness.

0241<figref idref="DRAWINGS">FIG. 24</figref> depicts a waveform plot <b>2400</b> of a corresponding output speech signal produced by SIE system <b>200</b> when the ambient background noise has increased to a level such that amplitude compression is applied to the far-end speech signal. Amplitude compression is used to allow for application of the full AVB gain necessary to reach the target SNR without digital saturation or clipping. As shown in plot <b>2400</b>, to accommodate the application of an increased AVB gain, the high-amplitude regions of the far-end speech signal have been compressed relative to the low-amplitude regions.

0242<figref idref="DRAWINGS">FIG. 25</figref> depicts a waveform plot <b>2500</b> of a corresponding output speech signal produced by SIE system <b>200</b> when the amount of amplitude compression applied due to background noise has increased to such a level that spectral shaping is applied to the far-end speech signal to preserve intelligibility. Spectral shaping operates to boost certain formants of the spectral envelope of the far-end speech signal above the near-end noise floor to make the far-end speech signal more intelligible. In one embodiment, the second and third formants of the far-end speech signal are boosted relative to the first formant since the second and third formants are more important from the perspective of speech intelligibility than the first formant.

0243A further example of the operation of SIE system <b>200</b> will now be described in reference to waveform plots shown in <figref idref="DRAWINGS">FIGS. 26-30</figref>. In particular, <figref idref="DRAWINGS">FIG. 26</figref> is a waveform plot <b>2600</b> of an exemplary far-end speech signal that may be received over a communication network and processed by SIE system <b>200</b>. <figref idref="DRAWINGS">FIG. 27</figref> is a waveform plot <b>2700</b> of exemplary ambient background noise present in the environment in which the telephony terminal that includes SIE system <b>200</b> is being used. <figref idref="DRAWINGS">FIG. 28</figref> is a waveform plot <b>2800</b> of an output speech signal produced by SIE system <b>200</b> responsive to processing the far-end speech signal depicted in plot <b>2600</b> of <figref idref="DRAWINGS">FIG. 26</figref> and the near-end background noise depicted in plot <b>2700</b> of <figref idref="DRAWINGS">FIG. 27</figref>. As shown in plot <b>2800</b>, SIE system <b>200</b> has boosted the portions of the far-end speech signal that coincide in time with the near-end background noise with the intent to achieve a minimum target far-end speech signal to near-end background noise ratio.

0244Assume that a user is using a telephony device that does not include SIE system <b>200</b> to play back the far-end speech signal plotted in <figref idref="DRAWINGS">FIG. 26</figref> in the context of the ambient background noise plotted in <figref idref="DRAWINGS">FIG. 27</figref>. Further assume that the telephony device includes a single loudspeaker that is housed in an ear bud which is inserted in the left ear of the user. In accordance with this example, <figref idref="DRAWINGS">FIG. 29</figref> depicts a first waveform plot <b>2902</b> that represents the audio content presented to the left ear of the user and a second waveform plot <b>2904</b> that represents the audio content presented to the right ear of the user. As shown in <figref idref="DRAWINGS">FIG. 29</figref>, the right ear of the user is presented with only the ambient background noise while the left ear of the user is presented with the far-end speech signal plus the ambient background noise in order to simulate and illustrate the experience of a user in a noisy environment with a telephony device on the left ear. In this example, much of the far-end speech will be unintelligible to the user due to the relative magnitude of the ambient background noise. It is noted that due to a seal between the ear bud and the left ear of the user, the magnitude of the ambient background noise presented to the left ear is less than that presented to the right.

0245In contrast, now assume that the user is using a telephony device that does include SIE system <b>200</b> to play back the far-end speech signal plotted in <figref idref="DRAWINGS">FIG. 26</figref> in the context of the ambient background noise plotted in <figref idref="DRAWINGS">FIG. 27</figref>. Further assume that the telephony device includes a single loudspeaker that is housed in an ear bud which is inserted in the left ear of the user. In accordance with this example, <figref idref="DRAWINGS">FIG. 30</figref> depicts a first waveform plot <b>3002</b> that represents the audio content presented to the left ear of the user and a second waveform plot <b>3004</b> that represents the audio content presented to the right ear of the user. As shown in <figref idref="DRAWINGS">FIG. 30</figref>, the right ear of the user is presented only the ambient background noise while the left ear of the user is presented with the SIE processed version of the far-end speech signal (shown in <figref idref="DRAWINGS">FIG. 28</figref>) plus the ambient background noise in order to simulate and illustrate the experience of a user in a noisy environment with an SIE enabled telephony device on the left ear. In this example, it can be seen from <figref idref="DRAWINGS">FIG. 30</figref> how the SIE is able to successfully process the speech signal so that it stands out from the background noise. Here again, it is noted that due to a seal between the ear bud and the left ear of the user, the magnitude of the ambient background noise presented to the left ear is less than that presented to the right. It should be noted that duration of the waveforms in <figref idref="DRAWINGS">FIG. 26</figref> through <figref idref="DRAWINGS">FIG. 30</figref> is approximately 9 minutes and 30 seconds, and the two highly noisy segments are each of approximately 3 minutes duration.

H. Example Computer System Implementations

0246It will be apparent to persons skilled in the relevant art(s) that various elements and features of the present invention, as described herein, may be implemented in hardware using analog and/or digital circuits, in software, through the execution of instructions by one or more general purpose or special-purpose processors, or as a combination of hardware and software.

0247The following description of a general purpose computer system is provided for the sake of completeness. Embodiments of the present invention can be implemented in hardware, or as a combination of software and hardware. Consequently, embodiments of the invention may be implemented in the environment of a computer system or other processing system. An example of such a computer system <b>3100</b> is shown in <figref idref="DRAWINGS">FIG. 31</figref>. All of the signal processing blocks depicted in <figref idref="DRAWINGS">FIGS. 2, 3 and 8-12</figref>, for example, can execute on one or more distinct computer systems <b>3100</b>. Furthermore, all of the steps of the flowcharts depicted in <figref idref="DRAWINGS">FIGS. 13-21</figref> can be implemented on one or more distinct computer systems <b>3100</b>.

0248Computer system <b>3100</b> includes one or more processors, such as processor <b>3104</b>. Processor <b>3104</b> can be a special purpose or a general purpose digital signal processor. Processor <b>3104</b> is connected to a communication infrastructure <b>3102</b> (for example, a bus or network). Various software implementations are described in terms of this exemplary computer system. After reading this description, it will become apparent to a person skilled in the relevant art(s) how to implement the invention using other computer systems and/or computer architectures.

0249Computer system <b>3100</b> also includes a main memory <b>3106</b>, preferably random access memory (RAM), and may also include a secondary memory <b>3120</b>. Secondary memory <b>3120</b> may include, for example, a hard disk drive <b>3122</b> and/or a removable storage drive <b>3124</b>, representing a floppy disk drive, a magnetic tape drive, an optical disk drive, or the like. Removable storage drive <b>3124</b> reads from and/or writes to a removable storage unit <b>3128</b> in a well known manner. Removable storage unit <b>3128</b> represents a floppy disk, magnetic tape, optical disk, or the like, which is read by and written to by removable storage drive <b>3124</b>. As will be appreciated by persons skilled in the relevant art(s), removable storage unit <b>3128</b> includes a computer usable storage medium having stored therein computer software and/or data.

0250In alternative implementations, secondary memory <b>3120</b> may include other similar means for allowing computer programs or other instructions to be loaded into computer system <b>3100</b>. Such means may include, for example, a removable storage unit <b>3130</b> and an interface <b>3126</b>. Examples of such means may include a program cartridge and cartridge interface (such as that found in video game devices), a removable memory chip (such as an EPROM, or PROM) and associated socket, and other removable storage units <b>3130</b> and interfaces <b>3126</b> which allow software and data to be transferred from removable storage unit <b>3130</b> to computer system <b>3100</b>.

0251Computer system <b>3100</b> may also include a communications interface <b>3140</b>. Communications interface <b>3140</b> allows software and data to be transferred between computer system <b>3100</b> and external devices. Examples of communications interface <b>3140</b> may include a modem, a network interface (such as an Ethernet card), a communications port, a PCMCIA slot and card, etc. Software and data transferred via communications interface <b>3140</b> are in the form of signals which may be electronic, electromagnetic, optical, or other signals capable of being received by communications interface <b>3140</b>. These signals are provided to communications interface <b>3140</b> via a communications path <b>3142</b>. Communications path <b>3142</b> carries signals and may be implemented using wire or cable, fiber optics, a phone line, a cellular phone link, an RF link and other communications channels.

0252As used herein, the terms “computer program medium” and “computer usable medium” are used to generally refer to media such as removable storage units <b>3128</b> and <b>3130</b> or a hard disk installed in hard disk drive <b>3122</b>. These computer program products are means for providing software to computer system <b>3100</b>.

0253Computer programs (also called computer control logic) are stored in main memory <b>3106</b> and/or secondary memory <b>3120</b>. Computer programs may also be received via communications interface <b>3140</b>. Such computer programs, when executed, enable the computer system <b>3100</b> to implement the present invention as discussed herein. In particular, the computer programs, when executed, enable processor <b>3100</b> to implement the processes of the present invention, such as any of the methods described herein. Accordingly, such computer programs represent controllers of the computer system <b>3100</b>. Where the invention is implemented using software, the software may be stored in a computer program product and loaded into computer system <b>3100</b> using removable storage drive <b>3124</b>, interface <b>3126</b>, or communications interface <b>3140</b>.

0254In another embodiment, features of the invention are implemented primarily in hardware using, for example, hardware components such as application-specific integrated circuits (ASICs) and gate arrays. Implementation of a hardware state machine so as to perform the functions described herein will also be apparent to persons skilled in the relevant art(s).

I. Conclusion

0255While various embodiments of the present invention have been described above, it should be understood that they have been presented by way of example, and not limitation. It will be apparent to persons skilled in the relevant art that various changes in form and detail can be made therein without departing from the spirit and scope of the invention. For example, although embodiments of the present invention are described herein as operating within the context of a telephony terminal, the present invention is not so limited and embodiments of the present invention may be implemented in any device capable of processing an audio signal for playback in the presence of background noise. Furthermore, the processing of an audio signal for playback as described herein may encompass processing the audio signal for immediate playback, processing the audio signal for storage followed by subsequent retrieval and playback, processing the audio signal for playback by the same device on which such processing occurs, or processing the audio signal for transmission to and playback by a different device.

0256The present invention has been described above with the aid of functional building blocks and method steps illustrating the performance of specified functions and relationships thereof. The boundaries of these functional building blocks and method steps have been arbitrarily defined herein for the convenience of the description. Alternate boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Any such alternate boundaries are thus within the scope and spirit of the claimed invention. One skilled in the art will recognize that these functional building blocks can be implemented by discrete components, application specific integrated circuits, processors executing appropriate software and the like or any combination thereof. Thus, the breadth and scope of the present invention should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.